A hybrid evolutionary approach for heterogeneous multiprocessor scheduling

Size: px
Start display at page:

Download "A hybrid evolutionary approach for heterogeneous multiprocessor scheduling"

Transcription

1 Soft Comput DOI 0.00/s FOCUS A hybrid evolutionary approach for heterogeneous multiprocessor scheduling C. K. Goh E. J. Teoh K. C. Tan Springer-Verlag 008 Abstract This article investigates the assignment of tasks with interdependencies in a heterogeneous multiprocessor environment; specific to this problem, task execution time varies depending on the nature of the tasks as well as with the processing element assigned. The solution to this heterogeneous multiprocessor scheduling problem involves the optimization of complete task assignments and processing order between the assigned processors to arrive at a minimum makespan, subject to a precedence constraint. To solve an NP-hard combinatorial optimization problem, as is typified by this problem, this paper presents a hybrid evolutionary algorithm that incorporates two local search heuristics, which exploit the intrinsic structure of the solution, as well as through the use of specialized genetic operators to promote exploration of the search space. The effectiveness and contribution of the proposed features are subsequently validated on a set of benchmark problems characterized by different degrees of communication times, task, and processor heterogeneities. Preliminary results from simulations demonstrate the effectiveness of the proposed algorithm in finding useful schedule sets based on the set of new benchmark problems. Keywords Multiprocessor scheduling Heterogeneous Hybrid evolutionary algorithm Local search Precedence C. K. Goh (B) Spintronics, Media and Interface Division, Data Storage Institute, DSI Building, Engineering Drive, Singapore 608, Singapore E. J. Teoh K. C. Tan Department of Electrical and Computer Engineering, National University of Singapore, 4 Engineering Drive, Singapore 6, Singapore Introduction The multiprocessor scheduling problem is a broad category of a class of combinatorial optimization problems in which an originally large problem is broken down into smaller tasks. These smaller, partitioned tasks then require suitable assignment to the individual processing units of a multiprocessor system or processing elements (PE) to be solved. To obtain solutions for optimal schedules in such systems, it has been shown that the problem is NP-hard for the general case (Garey and Johnson ; Kasahara and Narita 84; Lewis and El-Rewini ; Papadimitriou and Yannakakis 0). The underlying motivation for this problem is quite significant, considering the emergence of computer programs with increasingly higher computational requirements and algorithmic complexity. These factors have necessitated the need for parallel PE in a multi-computer environment, which in turn has seen the increasing need for task allocation to be optimally distributed in a suitable manner to these individual processing units. A typical program can usually be decomposed into a set of smaller tasks, similar to a divide-and-conquer approach. These smaller tasks almost always have dependencies, and hence precedence requirements in that the results of another set of tasks are required before a particular task can be executed. The critical aim of a scheduler is thus to assign partitioned tasks to available processors in a manner such that () the requirements (or constraints) of precedence between these tasks are met and () the resulting overall length of time required to execute the entire program, the schedule length or makespan, is minimized (Wu et al. 004). To complicate matters, the scheduling of tasks becomes more challenging when communication delays are accounted for. A multiprocessor scheduling problem can be categorized into different classes based on the characteristics of the

2 C. K. Goh et al. problem, the tasks to be scheduled, the multiprocessor system, as well as the availability of a priori information regarding the processing time (El-Rewini et al. 4; Kwok and Ahmad, ). Typically the PE constituting a multi-computer environment can be of the same capability (this is known as a homogenous environment) or of a different capability (this is known as a heterogeneous environment) this paper is focused on the latter. Presently, there are numerous methods and approaches which have been developed and subsequently applied to the multiprocessor scheduling problem, typically using a deterministic approach. El-Rewini et al.(4) provides a fairly comprehensive taxonomy of how scheduling problems can be categorized, and highlights the key differences that distinguishes one class from the next. Further to this, in Kwok and Ahmad (, ), present a wide-ranging overview and classification of scheduling algorithms, particularly focusing on deterministic and static scheduling problems. Most of the present techniques are based on heuristics (Kruatrachue and Lewis 8; Macey and Zomaya 8) that are not only greedy in nature but also capable of solving certain instances of the scheduling problem efficiently. With that in mind, the approach proposed here is largely inspired by developments in computational intelligence: evolutionary algorithms (EAs) are a class of stochastic global optimization techniques that has been gaining significant attention from researchers in many fields and it has also been applied to solve the heterogenous multiprocessor scheduling optimization problem (Ritchie and Levine 004; Zhong et al. 004). While EAs are excellent global search algorithms, it is known that they can take a relatively long time to locate the local optimum in the region of convergence (Ong et al. 006). On the other hand, local search heuristics are capable of locating the optimum quickly but are prone to local optimal traps. Therefore, EAs are often hybridized with local search heuristics to maintain a balance between exploration and exploitation, which is crucial to the success of search and optimization processes (Burke et al. 00; Franca et al. 00; Ishibuchi et al. 00; Merz and Freisleben 000; Ong and Keane 004; Tang et al. 00; Zhou et al. 00). Multiprocessor systems have also been exploited to improve EA performance (Lim et al. 00). This paper attempts to present a new hybrid evolutionary algorithm (HEA) for solving the above heterogeneous multiprocessor scheduling problem. The proposed algorithm incorporates two local search operators, based on firstly, list scheduling and secondly, task duplication; both methods attempt to exploit the intrinsic structure of the scheduling problem. Unlike existing evolutionary approaches used to solve the heterogeneous multiprocessor scheduling problem, the proposed HEA also implements a variable length chromosome which preserves the precedence relations, a PE schedule crossover which facilitates the exchange of good schedules assigned to the individual processors as well as specialized mutation operators to improve the diversity of the evolving population. This paper is organized as follows: Sect. gives an overview of existing works as well as the problem formulation of the heterogeneous multiprocessor scheduling problem. Section presents the various features of the proposed HEA including the local search heuristics and specialized genetic operators as well as the algorithmic flow. Section 4 presents the extensive simulation results and analysis of the proposed algorithm. Conclusions are then drawn in Sect.. Background information. Overview of existing works Multiprocessor scheduling based on methods motivated by evolutionary computation approaches have been the focus of many research works over the last decade. Here we offer a brief, non-exhaustive overview of similar works that have motivated our interests and research. Ahmad and Kwok (8) proposed a task duplication approach (together with a review and comparison of some similar algorithms) to mitigate the expensive communication overhead in interprocessor communications that is required when executing dependent tasks on multiple processors. In a similar manner, (Baskiyar and Dickinson 00) addresses static scheduling of a directed a-cyclic task graph (DAG) on a heterogeneous, bounded set of distributed processors to minimize the makespan, also based on a task duplication approach. Most of the present techniques are based on heuristics that are capable of solving only certain instances of the scheduling problem efficiently. However, the scheduling of tasks with communication overheads and dependencies are gaining increasing attention from researchers. Here, we investigate an alternative paradigm, based on biologically inspired algorithms, to efficiently solve the scheduling problem without the need to apply any restricting assumptions. Aside from the above, other works in the literature have used EAs to determine task priorities based on list scheduling techniques. List scheduling heuristic (LSH) is an approach involving the assignment of a priority to each task to be scheduled within a list, which is then subsequently sorted in decreasing task priority. The task with the highest priority in the unscheduled task list is typically assigned to the first available processor and then removed from the list. If there are more than one task being assigned the same priority level, selection from among the candidate tasks is typically done randomly. This conventional approach will be applied in the comparative study conducted in this paper. On the other hand, an alternative approach would be to use EAs to directly evolve task assignment and order in

3 A hybrid evolutionary approach for HMPS processors. Hou et al. (4) used an EA to evolve candidate solutions, or individuals that in turn consist of multiple lists, with each list representing the tasks assigned to one processor; the authors restrict the explorable design space in order to avoid invalid solutions. However in their proposed approach, the authors consider only homogeneous multiprocessor systems. Consequently, the crossover operation then exchanges tasks between corresponding processors from two different individuals, after which the mutation operator then exchanges these tasks within a single individual. Overall, this approach restricts the actions of genetic operators to ensure the validity of evolved individuals. However, such an approach would mean that some parts of the search space may be unreachable by the algorithm. Correa et al. () subsequently claims to improve upon Hou s original approach to circumvent this problem, and allow the entire search space to be explored. In Kwok and Ahmad (), proposed a coarse-grained parallel genetic algorithm (GA) together with a heuristical list scheduling method, where candidate solutions are vectors of length n, with n being the number of tasks to be scheduled. The elements of a vector represent the tasks themselves and the order of the tasks gives the relative task priorities. A number of order-based crossover operators are presented and a mutation operator is used to perform random swapping of tasks. In Dhodi et al. (), proposed a Problem Space Genetic Algorithm (PSGA) for datapath synthesis. The problem itself is modified by the EA and subsequently transformed into solution space by means of a heuristic, thus avoiding infeasible solutions. Blickle et al. (6) use an EA to perform allocation and binding on a system level. Scheduling is achieved in a separate step. The authors use multichromosomal individuals to encode the problem and to subsequently guide repair heuristics in parallel. Tsuchiya et al. (8) proposed an approach in which a GA scheduler allows task duplication where a single task may be assigned to multiple processors. Alternatively, Zomaya et al. () incorporate heuristics in the generation of the initial population of an EA and perform a thorough study of how GA performance varies with changing parameter settings. Wu et al. (004) claims that an EA-based approach achieves good performance on most of the problems applied. They also suggest that GAs appear to be the most flexible algorithm for heterogeneous systems because heterogeneous processors make it more difficult for list scheduling algorithms to accurately estimate task priority. An alternative approach, motivated by ant colony optimization (ACO), is developed by Ritchie and Levine (004). When combined with local and tabu search, the ACO-based algorithm is able to find shorter schedules on a few benchmark problems. ACO, as the authors also claim, has been shown to be a successful strategy for problems related to scheduling jobs in a heterogeneous computing environment. This approach was only tested in solving a scheduling problem in a static environment for independent jobs.. Heterogeneous multiprocessor scheduling problem Technological advancements have led to the development of large scale parallel and distributed systems for a large range of applications. However, applications are only able to exploit parallelism when their parts do not wait for data longer than necessary. This necessitates appropriate scheduling strategies, which are able to control access to processing resources, as well as scheduling strategies, which control execution of these parallel application modules. Thus, it is not surprising that the focus of research in this area has been on the efficiency and effectiveness of scheduling algorithms. There are increasing concerns that comparative studies performed are not adequate to evaluate the true abilities of the algorithms under test. Addressing the issue of data set generation, Hall and Posner (00) presented a set of guidelines on how data sets should be generated for the evaluation of the various scheduling algorithms. In order to generate a set of good test problems, the researcher must consider:. the purpose of the experiment,. tests performed should be comparable,. unintended bias that can skew the test results, and 4. the reproducibility of the generation scheme. Further, Hall and Posner also state that the generation scheme should have properties such as variety, practical relevance, scale and size invariance, regularity, describability, efficiency, and parsimony. Kwok and Ahmad () presented a suite of five different benchmark graphs. The proposed sets are peer set graphs (PSG), random graphs with optimal solutions using branch-and-bound (RGBOS), random graphs with predetermined optimal schedules (RGPOS), random graphs with no known optimal schedules (RGNOS), and traced graphs (TG). RGPOS is probably the most interesting set of task graphs in the sense that they are generated based on a set of pre-determined solutions. To our knowledge, this is the first instance of such generation scheme for a multiprocessor scheduling problem with communication delay. RGBOS also have a set of optimal solutions, which are determined using the A algorithm (Ahmad and Kwok 8). The A algorithm is a search heuristic, which incrementally searches all paths from the starting point until it finds the shortest path to a goal. PSG is a collection of task graphs used by various researchers. RGNOS consists of large scale randomly generated task graphs while TG represent real-world applications. Coll et al. (00) considered the issue of generating benchmark test sets for heterogeneous systems. The degree of heterogeneity between different processors is defined by a

4 C. K. Goh et al. processor power ratio (PPR), which represents the relative speeds between processors. In addition, they considered the different precedence relationships based on the specific nature of the task to be processed. More recently, Davidovic and Crainic (00) proposed a set of benchmark problems modeling homogeneous systems with communication delays. Based on the criteria proposed by Hall and Posner (00), they proposed two sets of task graphs. Similar to Kwok and Ahmad (), one of the proposed sets is generated based on some pre-determined desired solution. However, Davidovic and Crainic provide a much higher degree of control, allowing parameters such as dependency densities to be changed... Problem formulation The multiprocessor scheduling problem can be simply stated as follows: Assuming there are n tasks that have to be executed on m processors where and when should each task be executed, such that some performance measure(s) is (are) optimized? The task of the scheduling algorithm is to ultimately minimize a given cost function of time. The objective function used in this paper is defined as: { } F = min max T f (v i ) () i=,...,n where T f (v i ) denotes the time for the complete execution of task v i. The goal of task assignment/mapping is to determine an assignment of tasks to processors and an order in which tasks are executed to optimize some performance measures. Often, the assignment process should aim to minimize the total cost of executing the programs. An optimal assignment determines both the allocation (identifying specific processor to run certain modules) and the schedule (execution order) of each task. A task in turn, is a collection of instructions, procedures or subroutines, possibly together with some data. Each task is assumed to be immutable. While distributing the tasks to parallel PEs is not difficult, introducing dependencies between the tasks causes degradation of the overall system performance. There are bindings or linkages between some pairs of tasks (we call these dependencies) since a procedure in one task may wish to () transfer control to another procedure in a different task or () access data contained/produced in a different task. It should be noted that these tasks only incur a communication delay when they are assigned to different PEs. It is, thus, important to make the assumption that the cost of executing tasks on different processors and the cost of the communication delay are known in advance. As to why deterministic scheduling is considered, we are inclined to believe that a priori efforts must be devoted to analyze data from machine manufacturers and accumulate actual experience from running smaller programs on fewer processors. Such efforts are fully justified especially if repeated deterministic or production runs of important large programs will be run on large parallel systems where termination and successful results are expected. In fact, it is precisely for these production runs that the effort of optimizing assignment is justified in the first place. The duration of each task is known as well as precedence relations among tasks, i.e. which tasks should be completed before some others can begin. In addition, if dependent tasks are executed on different processors, data transferring times or communication delays that are given in advance are also considered. These latencies also include memory access and synchronization delays. To further include realism into our problem model, we also consider a heterogeneous system, that is a multiprocessor environment consisting of processors with different capabilities. Moreover, we only consider a non-preemptive system, that is, each PE will complete the processing of each task that is assigned to it. Essentially, this means that PEs will not suspend its processing to take on another task... Problem generator In order to verify the efficacy of our proposed approach, a set of problems are needed for the experimental study. This is achieved via the construction of a benchmark problem generator, which produces a representative problem of a certain complexity based upon a set of input parameters. These test problems are in turn used as the input problem to the task scheduler. Having said that, there are four key components in a task scheduler: the parallel program of interrelated tasks, the target machine (model), the generated schedule, and the performance criterion. Previous works on task scheduling with dependencies usually use a graph representation for either the tasks of the parallel program and the computer model, or both. In an actual multiprocessor computing system, particularly those consisting of heterogeneous elements, the running time of a particular job is not the sole or primary factor to be considered when scheduling jobs. An equally important consideration is the time that it takes to migrate the executables and its associated data from one processor to the next. Braun et al. (00) defined three types of heterogeneity: task heterogeneity, machine heterogeneity and consistency. Task heterogeneity is defined as the amount of variance possible among the execution times of the jobs. Machine heterogeneity, on the other hand, represents the variation of the running time of a particular job across the processors. Lastly, consistency can be categorized as either: consistent,

5 A hybrid evolutionary approach for HMPS inconsistent and semi-consistent. A system is said to be consistent if for a processor A that executes a job C faster than another processor B, then A will execute all other jobs faster than B. A consistent system can therefore be seen as modeling a heterogeneous system in which the processors differ only in their processing speed. A semi-inconsistent system is made of elements from both consistent and inconsistent systems. Higher degrees of machine heterogeneity increase the complexity of the multiprocessor scheduling problem. This is because the scheduling algorithm now needs to account for the variation in the individual processor s capabilities, in that certain processors might be more suitable for certain tasks due to hardware or software configurations and compatibility. The multiprocessor system is made up of m processors with their own local memories. The system can have various degree of heterogeneity and the processors are connected via bi-directional links of equal capacity. Each processor has an I/O unit that allows for communication and processing to be performed simultaneously. We assume that there are no start-up costs for initiating each task and that input buffers have infinite capacity. A convenient representation for the partially ordered set of tasks is a directed acyclic graph (DAG), which is also known as a task (dependency) graph, where a directed edge e(p, j) between two tasks v p and v i specifies that task v p must be completed before v i can begin. These directed edges in a DAG correspond to the communication messages as well as precedence constraints between the tasks. We consider a node and a task to be equivalent. A task is a set of instructions that must be executed sequentially in the same processor. They are considered to be the smallest possible instruction set that cannot be broken up any further. Mathematically, node v p is a predecessor of node v i if a directed edge originates from v p and ends at v i. In a similar manner, node v s is a successor of node v i if a directed edge originating from v i and ending at v s exists. From a mathematical perspective, for any vertex v in the DAG, there is no non-empty directed path that starts and ends on v as such, for our multiprocessor task scheduling problem, DAGs are quite ideal models since it is not tractable for a vertex to have a path to itself; for example, if an edge v p v i indicates that v i is a part of v p, such a path would indicate that v p is a part of itself, which is impossible. The test sets that were artificially generated using our benchmark problem generator is based on this concept of DAGs. From a practical viewpoint, actual multiprocessor systems are immensely complicated combinations of hardware, software and network components and thus it is difficult to make equitable comparisons of the different approaches that have been used on various systems. In constructing these problems artificially, and in a random manner, the input variables essentially controls not only the size, but also the Table Description of inputs to task generator Parameter Description Values CCR Communication-to-computation ratio {0.,,.,} Mean proc Mean processing time {0} h pe Variance of processing time {0.,0.,0.} h t Degree of heterogeneity {0.,0.,0.} dpe Width of DAG {0.} dt Degree of dependency {0.,0.,0.} n Number of processors {} m Number of tasks {00} complexity of the generated test set. Specifically, these variables are:. the number of nodes/tasks,. the number of processors available,. the degree of network connectivity, 4. the communication-to-computation ratio average communication cost divided by its average computation cost in a multiprocessor system. A low CCR in a DAG can be considered as a computation-intensive application; on the other hand if CCR is high, it is a communicationintensive application,. the mean processing time the average processing time for all the available processors, 6. the variance of processing time how large the spread of processing time between the available processors,. the degree of heterogeneity how widely differing the capabilities of the processors are, i.e. processors have different execution time on same the task, 8. the degree of precedence/dependency relationship how many predecessor tasks that must be completed before a particular task can be executed. Having said that, the generator produces different test sets for a given set of input parameters. For similar set of parameter, different task problems are generated due to randomness. The input parameters to the generator are shown in Table, together with the associated range of values. The variance in the processing times of the difference tasks comes from h pe, i.e the mean processing time of the ith task is given by T mproc (v i ) = mean proc + h t mean proc U(, ). () where U(, ) denotes a random number sampled using uniform distribution. As mentioned before, each task may have different execution times on different processors. The actual processing time of the ith task on the jth processor is thus given by, T proc (v i, pe j ) = T mproc (v i ) + h pe mean proc U(, ). ()

6 C. K. Goh et al. Table Generated test sets Test set CCR Mean proc h pe h t dpe dt n m T T T T T T T T T T Using these inputs for the benchmark problem generator, sets of random DAGs were constructed to be used as the test bed problems in our experimental study. For our simulation study, ten test sets were generated using various combination of the above input parameters, and are listed in Table. While the standard multiprocessor scheduling problem is itself an NP-hard problem, additional factors such as communication delays and heterogeneity increase the complexity of the problem. Hence, due to the sheer number of potential solutions in the search space, scheduling becomes a complex task without the use of an effective search algorithm. These sets are classified in terms of the possible difficulties. Each test set consists of different test problems with different degrees of heterogeneity and dependencies. Here, we consider a total of ten test sets generated in this study, which differs in terms of degree of heterogeneity, density, and CCR. A higher CCR value penalizes dependencies which require transmission or passing of messages from one processor to the next, making it less optimal for inter-processor communication to occur. The variance of processing time and degree of heterogeneity affects the individual processing capabilities of each processor, thus making slower processors less likely to be assigned tasks, and biasing the utility of faster processors. Lastly, the degree of dependency affects the total latency of the makespan in that each processor would have to wait for its dependent tasks to finish execution. Hybrid evolutionary algorithm This section presents the HEA specifically designed to solve the heterogeneous multiprocessor scheduling problem by means of specialized genetic and local search operators. The procedure for generating the initial population is presented in Sect.. while Sect.. describes the structure of the variable-length chromosome used to encode the task schedule in the HEA. Sections. and.4 describe the specialized crossover and mutation operators used to explore the search space, respectively. Two local search heuristics that exploit the intrinsic structures of a heterogeneous multiprocessor scheduling problem solution are presented in Sect... Finally, the algorithmic flow of the HEA is presented in Sect..6.. Initialization The initial population is built using a random LSH, which ensures that the precedence relationships among the tasks are preserved. The initialization process starts with the assignment of priority to each task to be scheduled. In this paper, the priority of the ith task is simply the sum of the number of its parent tasks and their priorities as given below Pr Ti = P i + Pr Tj (4) j P i where P i is the set of parent tasks of the ith task. The list of task is then sorted in the order of increasing priority. This priority list is also used during the genetic processes to maintain the precedence requirements. Instead of assigning the tasks to the earliest available PE, the lowest priority task is assigned to the PEs randomly. The rationale is to provide the initial population with a wider range of diversity to start with.. Variable PE chromosome Evolutionary algorithms operates on a set of encoded parameters to explore the solution space, providing researchers with the flexibility to design an appropriate representation that fulfills some criteria such as ease of implementation or exploitation of the problem structure. For simplicity, the chromosome is often represented as a fixed-structure and the embedded variables are usually assumed to be independent and context insensitive. As mentioned before, the precedence relations among the tasks must be satisfied in the heterogeneous multiprocessor scheduling problem. In Braun et al. (00); Ritchie and Levine (004), the chromosome is a n-dimensional array denoting the n tasks to be allocated and the encoded variable in each element represents the PE scheduled to execute the associated task. While such an encoding scheme is simple to implement, it does not consider the order in which the various tasks are processed and the evolved schedules will not satisfy the precedence constraints. On the other hand, Wu et al. (004) considered a representation which encodes task-processor pairs and the order in which the pairs appear in the chromosome determines the order in which the tasks will be performed on each processor. This paper adopts a variable length chromosome which is illustrated in Fig.. In contrast to the mentioned works, this encoding scheme does not enforce a fixed number of PEs, i.e.

7 A hybrid evolutionary approach for HMPS Fig. Illustration of a the variable length chromosome and b the associated schedule (a) PEs used in the encoded solution PE PE PE4 PE6 (b) Processors 0 4 PE 4 6 PE PE Tasks to be executed by the associated PE PE4 PE 0 PE Schedule PE6 Tasks 4 the length of the chromosome varies with the actual number of PE utilized. For each of these PEs, there is an associated list of task assigned as well as the order of execution. Each of the task list will henceforth be denoted as PE schedule. When a task is scheduled to run before its predecessor tasks, which have been assigned to other PEs, the only problem is the long idle time incurred while waiting for all the predecessor tasks to be completed. On the other hand, if a task is scheduled to run before its predecessor tasks on the same PEs, then there is no way the task will ever be completed. The overall schedule is infeasible only if a task is scheduled to be executed before a predecessor task within a PE. This follows that it is sufficient to maintain a feasible overall schedule by ensuring the feasibility of each PE schedule. The precedence relations for the tasks executed in a PE can be easily preserved in the proposed scheme by maintaining the order of priority calculated at the beginning of the optimization process.. PE schedule crossover The crossover operation applied by most EAs to solve heterogeneous multiprocessor scheduling problem generally involve the swapping of random segments of tasks or processes between chromosomes, which do not preserve the quality of the different PE schedules. Descriptions of a number of ordered-based crossovers for combinatorial problems can also be found in Davis (), Eiben and Smith (00). However, these crossover operators are not applicable due to the unique structure of the proposed variable length chromosome. The proposed PE schedule crossover is motivated by the fact that the makespan of the multiprocessor schedule is dependent on the fitness of the constituent PE schedules. Since the chromosome encodes a separate list of tasks for each PE, it is intuitive to design a crossover which allows good PE schedules to be shared with other chromosomes in the evolving population. The operation of the crossover is illustrated in Fig.. In the PE schedule crossover, a random PE schedule from each parent is selected for crossover. In the case where one of the selected chromosomes has only one PE schedule, only a schedule associated with a different PE is selected and inserted from the other parent. The selected PE schedule of one parent will either be inserted into the other chromosome as a new schedule or replaces the original schedule of that particular PE, if it is present. Duplicated tasks are deleted while missing tasks are randomly inserted to the other original PE schedules. The new PE schedule will remain intact. To ensure the feasibility of chromosomes after the crossover, the priority list computed at the beginning of the evolutionary process is used to sort the task assigned to each PE in ascending order to preserve feasibility..4 Specialized mutation This paper applies three different specialized mutation operators to improve the diversity of evolving population. For every chromosome undergoing the mutation process, only one particular mutation operator is applied as shown by the pseudocode in Fig.. The main functionalities of the three mutation operators are summarized in Table. Similar to the PE Schedule crossover, each PE schedule is sorted based on the priority list at the end of the mutation operation.. Local search.. Partial list scheduling The optimality of the multiprocessor schedule is only as good as the last completion time of the task. The idea of partial list scheduling (PLS) is to split up the workload among the PEs with the best and worst completion times to improve the makespan. The first step in this heuristic is to select the appropriate PEs from which all tasks are extracted and placed in a list. These PEs are selected based on two criteria, either the

8 C. K. Goh et al. (a) Parent Parent (b) Child Child PE PE PE4 PE6 PE PE PE6 PE PE PE4 PE PE6 PE PE PE (c) Child Child PE PE PE4 PE PE6 PE PE PE Fig. Illustration of the PE schedule crossover for the various steps a selection of random PE schedule, b swapping of selected PE schedules, and c deletion of duplicates and random insertion of missing tasks to form child chromosome Mu tation Operation IF rand < mutation rate Select one mutation operator with equal probability IF Partial Exchange AND No. of PE Schedules > Perform Partial Exchange ELSEIF Schedule Merge AND No. of PE Schedules > Perform Schedule Merge ELSEIF Partial Split AND No. of PE Schedules < PE Perform Partial Split Sort task based on priority Fig. Pseudocode of the mutation operation PE has a completion time that is greater than the upper quartile or it s completion time is lower than the lower quartile of the PE completion times. In the next step, the extracted tasks are sorted based on their priorities determined at the start of the evolutionary process. The tasks, in the order of their priorities, are then assigned to the best possible processor, i.e. the one which allows the earliest start time considering inter-task communication (ITC). The new solution will be compared against the original and the better of the two will be retained... Duplication scheduling In multiprocessor scheduling with task interdependencies, some PEs will be idle during various time slots because some task require data from its parent tasks which are assigned to other processors. The idea of duplicating tasks in these idle time slots is to reduce the waiting and ITC delays incurred to reduce the makespan. The pseudocode of the duplication scheduling (DS) heuristic is shown in Fig. 4. The task duplication procedure is conducted iteratively every task in the order of its execution for each PE. The heuristic first determines the idle time which is the difference between the actual and earliest possible start time of the task. It then attempts to duplicate the parent tasks, in the order of their contribution to the delay, until the idle time is used up. The new solution will be compared against the original and the better of the two will be retained..6 Algorithmic flow The algorithmic flow of the HEA is shown in Fig.. The optimization process begins with the initialization of the

9 A hybrid evolutionary approach for HMPS Table Description of the mutation operation Operator Partial exchange Schedule merge Partial split Description The partial exchange operation involves a number of partial schedule exchanges. For each exchange, two PE schedules are randomly chosen and a segment of the selected schedules is then randomly selected and exchanged. In addition, a mechanism is in place such that no PE schedules will be selected twice in a particular partial exchange operation This operation concatenates the two PE schedules with the least number of tasks in the chromosome. Intuitively, this operation is not applicable to solutions with only one PE schedule This operation searches for the PE schedules with the most number of tasks, and breaks the schedule into two at a random point. After which, the upper segment of the divided schedule is assigned randomly to either an idle PE or inserted into the PE schedule with the least number of tasks Duplication S cheduling L ocal Search FOR All PE Schedules FOR All Task in PE Schedule Compute T idle before task execution Determine parents of task Sort parents in descending order of completion time FOR Parents Determine T exe required if d uplicated IF Execution time< T idle Duplicate parent Update T idle : T idle = T idle - T exe ELSE Break Sort task based on priority Evaluate new solution IF new solution is better than old solution Replace old solution YES Evaluate and Rank Solutions Update Archive Return Solution Start Build Initial Population Stopping Criteria met? No Evaluate and Rank Solutions Update Archive Perform Local Search Tournament Selection Local Search Criteria met? Yes No Fig. 4 Pseudocode of the duplication scheduling local search PE Schedule Crossover Mutation population based on the procedure described in Sect... After the initial evolving population is formed, all the chromosomes are evaluated and ranked according to their final execution time in the population. Following the ranking process, an archive population is updated. In this paper, an archive is applied to store all the best solutions found during the search. The archive maintains a fixed number of solutions and the updating process consists of a few steps. The evolving population and the archived solutions are first combined and all duplicate solutions are deleted. The remaining solutions in the combined population are then inserted into the archive in the order of increasing rank until the archive is filled. The binary tournament selection scheme is then performed on the archive. In the binary tournament selection, a pair of individuals is selected randomly from the archive. Thereafter, the selected pair of individuals will enter a tournament where the chromosome with the lower rank is selected for reproduction. This procedure is performed until the mating pool is filled to preserve the original population size. The genetic operators consist of the PE schedule crossover and Fig. Flowchart of HEA the three mutation operators presented in Sects..4 and., respectively. The PLS and DS are applied to the archive populations at a fixed interval, T LS, for better local exploitation in the evolutionary search. Different schemes for incorporating the two local search methods will be explored in Sect. 4.The evolution process is repeated until the stopping criterion is satisfied. 4 Simulation results and analysis This section presents the extensive simulation results and analysis of the proposed HEA. The simulations are implemented using Matlab on an Intel Pentium 4.8 GHz computer and the results shown are based on the final makespan value of the best archived solution. Thirty independent runs are performed for each of the test sets in order to obtain the statistical information, such as consistency and robustness of the

10 C. K. Goh et al. Table 4 Parameter setting for HEA Parameter Settings Populations Population size 0 Archive size 0 Chromosome Variable length chromosome Selection Binary tournament selection Crossover rate 0. Mutation rate 0. Evaluations 600 Local search frequency, T LS algorithms. The various parameter settings for the algorithm are listed in Table 4. The number of evaluations includes the evaluation of solutions from the main algorithmic cycle as well as the local search operations. Section 4. demonstrates the effectiveness of the proposed local search operators, as well as analyzes how the various settings of the local search heuristics will affect algorithmic performances. Section 4. investigates the impact of different problem characteristics on HEA performances and how it compares against conventional heuristics. 4. Effects of local search The HEA incorporates the local search heuristics in order to exploit local schedules in parallel with global evolutionary optimization. In this section, the dynamics and parameter settings of PLS and DS are examined. Note that T and T4 are used in the study here since it has been observed in previous works that ITC will have severe impact on schedule optimality. Six settings of HEA with various implementations of the local search operators are investigated as shown in Table. No local search is applied in setup while only one heuristic is applied for each solution undergoing local search for setups 6. In setup, either PLS or DS is randomly applied. Table Different case setups to examine contribution of the local search heuristics 4 6 PLS Random Yes Alternate Alternate DS Random Yes Alternate Alternate In the third and fourth setup, only one heuristic is applied. The asterisk ( ) in setup and setup 6 denotes which local search is activated first as they are alternately executed. The evolutionary trends of the makespan averaged over 0 runs for T and T4 are plotted in Fig. 6a, b. From the plots, it can be observed that the application of local search results in significant dips in the convergence trace, particularly in instances where DS is applied. Figure 6a, b distinctively demonstrate the effectiveness of local exploitation in the HEA as the five setups which incorporates local search performed better as compared to setup. The performances of setup, setup 4, setup, and setup 6 are comparable, although the combination of DS being activated first and PLS in setup 6 seems to have a slight edge for both problems. On the other hand, setup which activates PLS first has a slower convergence rate for both T and T4. The effectiveness of duplicating tasks in reducing overall completion time is also evident since the four settings of setup, setup, setup and setup 6, are able to find solutions with makespans that are significantly lower than those found without local search and by PLS only. Interestingly, the application of DS seems to have more impact on T with an average of 0% improvement as compared to % for T4 which has a more severe CCR restriction. Setup 6 will be used as the default setup for all subsequent experiments. 4.. Effect of local search frequency In general, there is a need to maintain a balance between exploration and exploitation. Therefore, experiments are also conducted to study the impact of local search frequency on (a) Setup Setup Setup Setup4 Setup Setup Evaluation (b) Setup Setup Setup Setup4 Setup Setup Evaluation Fig. 6 Evolutionary trend of the six setups for a T and b T4

11 A hybrid evolutionary approach for HMPS (a) (b) Evaluation Evaluation Fig. Effects of various T LS settings for a T and b T (a) (b) Archive only Population only Both Setups Archive only Population only Both Setups Fig. 8 of HEA with different individual local search selection schemes for a T and b T4 the performance of the HEA. Apart from the original setting of applying local search at T LS =, four other settings where local search is applied in every generation and at intervals of T LS ={,, 0} generations, respectively, are used in the test. Thirty simulation runs of each of the five settings were performed and the convergence traces are plotted in Fig. a, b. It should be noted that the maximum number of evaluations is maintained at 400 for all simulations, i.e. increasing the frequency of local search reduces the number of generations. From the figures, it can be observed that the convergence speed increases with decreasing T LS. While it is expected that increasing local search frequency will improve convergence speeds, there is always a risk of yielding local optimum solutions due to the lack of sufficient exploration. Nonetheless, we note that a well-designed LSH is capable of achieving schedules within % of the optimal solution. By comparing the results achieved by the HEA and the conventional heuristics which will be shown in the Sect. 4., it is thus unlikely that the HEA is trapped in a local optimal. This is probably due to the global exploration capability of the HEA. 4.. Effect of individual selection Apart from T LS, another factor that will influence the effectiveness of the local search process is the selection of individuals. In the preceding sections, only archived individuals are exploited by the heuristics. In this section, two other methods of individual selection are investigated. In the second approach, only individuals from the evolving population will undergo local search while random individuals are selected via tournament selection from the archive and evolving population in the third approach. The simulation results are summarized in the form of boxplots in Fig. 8a, b. From the figures, it can be observed that performing local search on archived solutions have an edge as compared to exploiting the evolving population only. On the other hand, the original approach of exploiting the archive only is comparable to the third approach which exploits a selected set of archived and evolving population individuals. The KS-test is also compared and it showed that the only the second method is statistically different from the other methods.

12 C. K. Goh et al. Table 6 of HEA with and without the various genetic operators for T and T4 T T4 First quartile HEA Median Third quartile First quartile Crossover only Median Third quartile First quartile Mutation only Median Third quartile The best result in highlighted in bold 4. Effects of genetic operators This section examines the contribution of the specialized genetic operators to the performance of HEA on the problems of T and T4. In order to assess the effects of the PE schedule crossover and the mutation operators, simulations for two different setups of HEA are conducted. Specifically, the first setup incorporates only the PE schedule crossover while only the mutation operator is implemented in the second setup. The simulation results are summarized in Table 6. The results indicate a deterioration of algorithmic performance when either the crossover or mutation operator is removed. Nonetheless, it can be observed that the crossover has a greater impact on algorithmic performance indicating the importance of exchanging PE schedules between individuals. The KS-test conducted also showed that HEA is statistically better than HEA with mutation operator only. 4. Investigation of other test problems In order to examine the effectiveness of HEA, a comparative study with conventional LSH and duplication scheduling heuristic (DSH) (Kruatrachue and Lewis 8) is carried out based upon the ten test problems described earlier. LSH has been described earlier in Sect... DSH is an instantiation of the LSH with the task duplication described in Sect... Specifically, in DSH, parent task are assigned into idle slots whenever possible after all tasks are assigned using LSH. As before, 0 simulation runs are conducted for all test problems and the results are summarized in Table. LSH and DSH are deterministic heuristics and only one solution is produced for each problem. As noted before in Sect. 4., the effectiveness of task duplication is evident by comparing the performances between LSH and DSH. The difference between the two conventional heuristics becomes even more apparent as the CCR or degree of heterogeneity increases. On the other hand, the Table Simulation results of LSH, DSH and HEA for the various benchmark problems LSH DSH First HEA Third quartile Median quartile T T T T T T T T T T HEA outperforms both heuristics for all test problems. With the exception of T, it can be observed from Table that the third quartile makespan value attained by HEA is much lower as compared to LSH and DSH for the benchmark problems. This also implies that the HEA is capable of evolving good schedules consistently. In order to analyze the impact of the various problem parameters, the performance trend over the different settings is plotted in Fig. a, d. In general, increasing the degree of CCR, precedence and task heterogeneity result in higher makespans for all algorithms. Nonetheless, it can be observed that the problem and algorithmic performances have different sensitivities toward these parameters. For instance, total execution time seems to vary almost linearly with CCR and task heterogeneity. As CCR increases beyond a certain threshold, we can expect that solutions which employ fewer PEs or, at least, concentrate the workload on a few core PEs to become more desirable. On the other hand, the initial increment in the degree of precedence relation from to 0% leads to a sharp increase in makespan. This is probably due to the subsequent increase in waiting time before a task can be executed. However, it can be seen from Fig. b that such an effect seems to saturate as the degree of precedence is further increased to %. Interestingly, we can observe from Fig. c that increasing PE heterogeneity actually improves the makespan. This behavior can be attributed to the PEs that can be either very efficient or inefficient with certain tasks. As a result, HEA is able to exploit such a problem characteristic to generate schedules that are much better compared to DSH and LSH. Conclusion Task scheduling in a multiprocessor system is an NP-hard problem that is critical in distinguishing the performance of a

13 A hybrid evolutionary approach for HMPS (a) 4 40 (b) CCR Degree of Precedence (%) (c) (d) Degree of PE Het (%) Degree of Task Het (%) Fig. Performance trend of HEA (open circle), LHS (open inverted triangle) and DHS (open square) for various degrees of a CCR (results for T, T, T and T4), b precedence (results for T, T and T6), c PE heterogeneity (results for T, T and T8) and d task heterogeneity (results for T, T and T0) multiprocessor system over a single processor system. However, the fact that tasks are located on different PEs mean that additional overheads are incurred. Furthermore, the aim of minimizing time is only a single objective practical requirements require that other measures of cost are minimized as well. In this article, we proposed a HEA specifically designed to solve the heterogeneous multiprocessor scheduling problem by means of a variable-length chromosome, as well as specialized genetic and local search operators. The starting population is initialized using a random LSH to preserve the precedence relationships between the tasks. The evolutionary process is driven two primary variation operator a schedule crossover and three variants of the mutation operator partial exchange, schedule merge, and schedule split; the local search operators on the other hand consists of a partial list scheduling and duplication scheduling approach. In presenting our results based on a fairly extensive simulation study, we showed that, the proposed genetic operators, when coupled with the local search operators performed better than in the case where any one of the operators were omitted. By incorporating the local search operators, particularly duplication local search into the overall proposed algorithm, convergence time, as expected, was shown to decrease. This observation becomes more evident on test sets where the CCR is smaller. References Ahmad I, Kwok YK (8) Optimal and near-optimal allocation of precedence-constrained tasks to parallel processors: defying the high complexity using effective search techniques. In: Proceedings of 8 international conference on parallel processing, pp 4 4 Ahmad I, Kwok YK (8) On exploiting task duplication in parallel program scheduling. IEEE Trans Parallel Distrib Syst (): 8 8 Baskiyar S, Dickinson C (00) Scheduling directed a-cyclic task graphs on a bounded set of heterogeneous processors using task duplication. J Parallel Distrib Comput 6(8): Blickle T, Teich J, Thiele L (6) System level synthesis using evolutionary algorithms, TIK-Report, Nr. 6 Braun TD, Siegel HJ, Beck N, Boloni LL, Maheswaran M, Reuther AI, Robertson JP, Theys MD, Yao B, Hensgen D, Freund RF (00)

14 C. K. Goh et al. A comparison of eleven static heuristics for mapping a class of independent tasks onto heterogeneous distributed computing systems. J Parallel Distrib Comput 6(6):80 8 Burke EK, Cowling P, De Causmaecker P (00) A memetic approach to the nurse rostering problem. Appl Intell (): 4 Coll PE, Ribeiro CC, de Sousa CC (00) Test instances for scheduling unrelated processors under precedence constraints. inf.pucrio.br/celso/grupo/readme.ps Correa RC, Ferreira A, Rebreyend P () Scheduling multiprocessor tasks with genetic algorithms. IEEE Trans Parallel Distrib Syst 0(8):8 8 Davidovic T, Crainic TG (00) New benchmarks for static task scheduling on homogenous multiprocessor systems with communication delays, Publication CRT, 00-04, Centre de Recherche sur les Transports, Universite de Montreal, pp 6 Davis L () Handbook of genetic algorithms. Van Nostrand Reinhold, London Dhodi MK, Hielscher EH, Storer RH, Bhasker J () Datapath synthesis using a problem space genetic algorithm. IEEE Trans CAD 4(8):4 44 Eiben AE, Smith JE (00) Introduction to evolutionary computing. Springer, New York El-Rewini H, Lewis TG, Ali HH (4) Task scheduling in parallel and distributed systems. Prentice Hall, Englewood Cliffs Franca PM, Mendes A, Moscato P (00) A memetic algorithm for the total tardiness single machine scheduling problem. Eur J Oper Res ():4 4 Garey MR, Johnson DS () Computers and intractability, a guide to the theory of NP-completeness. W.H. Freeman and Co., San Francisco Hall NG, Posner ME (00) Generating experimental data for computational testing with machine scheduling applications. Oper Res 4:84 86 Hou ES, Ansari N, Ren H (4) A genetic algorithm for multiprocessor scheduling. IEEE Trans Parallel Distrib Syst (): 0 Ishibuchi H, Yoshida T, Murata T (00) Balance between genetic search and local search in memetic algorithms for multiobjective permutation flowshop scheduling. IEEE Trans Evol Comput ():04 Kasahara H, Narita S (84) Practical multiprocessor scheduling algorithms for efficient parallel processing. IEEE Trans Comput ():0 0 Kruatrachue B, Lewis TG (8) Duplication scheduling heuristic, a new precedence task scheduler for parallel systems, Technical Report 8-60-, Oregon State University Kwok Y, Ahmad I () Efficient scheduling of arbitrary task graphs to multiprocessors using a parallel genetic algorithm. J Parallel Distrib Comput 4():8 Kwok Y, Ahmad I () Static scheduling algorithms for allocating directed task graphs to multiprocessors. ACM Comput Surv (4):406 4 Lewis TG, El-Rewini H () Introduction to parallel computing. Prentice Hall, New York Lim D, Ong YS, Jin Y, Sendhoff B, Lee BS (00) Efficient hierarchical parallel genetic algorithm using grid computing. In: Future generation computer systems: the international journal of grid computing: theory, methods and applications, pp Macey BS, Zomaya AY (8) A performance evaluation of CP list scheduling heuristics for communication intensive task graphs. In: Proceedings of the joint th international parallel processing symposium and ninth symposium on parallel and distributed programming, pp 8 4 Merz P, Freisleben B (000) Fitness landscape analysis and memetic algorithms for the quadratic assignment problem. IEEE Trans Evol Comput 4(4): Ong YS, Keane AJ (004) Meta-Lamarckian learning in memetic algorithms. IEEE Trans Evol Comput 8(): 0 Ong YS, Lim MH, Zhu N, Wong KW (006) Classification of adaptive memetic algorithms: a comparative study. IEEE Trans Syst Man Cybern B 6():4 Papadimitriou C, Yannakakis M (0) Toward an architecture independent analysis of parallel algorithms. SIAM J Comput : 8 Ritchie G, Levine J (004) A hybrid ant algorithm for scheduling independent jobs in heterogeneous computing environments. In: Proceedings of the rd workshop of the UK planning and scheduling special interest group Tang J, Lim MH, Ong YS (00) Diversity-adaptive parallel memetic algorithm for solving large scale combinatorial optimization problems. Soft Comput ():8 888 Tsuchiya T, Osada T, Kikuno T (8) Genetic-based multiprocessor scheduling using task duplication. Microprocessors Microsyst : 0 Wu AS, Yu H, Jin S, Lin KC, Schiavone G (004) An incremental genetic algorithm approach to multiprocessor scheduling. IEEE Trans Parallel Distrib Syst ():84 84 Zhou Z, Ong YS, Lim MH, Lee BS (00) Memetic algorithm using multi-surrogates for computationally expensive optimization problems. Soft Comput (0): Zhong YW, Yang JG, Qi HN (004) A hybrid genetic algorithm for task scheduling in heterogeneous computing systems. In: Proceedings of the third international conference on machine learning and cybernetics, pp Zomaya AY, Ward C, Macey B () Genetic scheduling for parallel processor systems: comparative studies and performance issues. IEEE Trans Parallel Distrib Syst 0(8): 8

THE emergence of computer programs with increasingly

THE emergence of computer programs with increasingly The Eighth International Conference on Intelligent Technologies (InTech 0) Sydney, Australia, - December 00 A Hybrid Evolutionary Approach for Heterogeneous Multiprocessor scheduling K. C. Tan, C. K. Goh,

More information

A Comparison of General Approaches to Multiprocessor Scheduling

A Comparison of General Approaches to Multiprocessor Scheduling A Comparison of General Approaches to Multiprocessor Scheduling Jing-Chiou Liou AT&T Laboratories Middletown, NJ 0778, USA jing@jolt.mt.att.com Michael A. Palis Department of Computer Science Rutgers University

More information

Evaluation of Different Task Scheduling Policies in Multi-Core Systems with Reconfigurable Hardware

Evaluation of Different Task Scheduling Policies in Multi-Core Systems with Reconfigurable Hardware Evaluation of Different Task Scheduling Policies in Multi-Core Systems with Reconfigurable Hardware Mahyar Shahsavari, Zaid Al-Ars, Koen Bertels,1, Computer Engineering Group, Software & Computer Technology

More information

A Service Revenue-oriented Task Scheduling Model of Cloud Computing

A Service Revenue-oriented Task Scheduling Model of Cloud Computing Journal of Information & Computational Science 10:10 (2013) 3153 3161 July 1, 2013 Available at http://www.joics.com A Service Revenue-oriented Task Scheduling Model of Cloud Computing Jianguang Deng a,b,,

More information

Memory Allocation Technique for Segregated Free List Based on Genetic Algorithm

Memory Allocation Technique for Segregated Free List Based on Genetic Algorithm Journal of Al-Nahrain University Vol.15 (2), June, 2012, pp.161-168 Science Memory Allocation Technique for Segregated Free List Based on Genetic Algorithm Manal F. Younis Computer Department, College

More information

Population-based Metaheuristics for Tasks Scheduling in Heterogeneous Distributed Systems

Population-based Metaheuristics for Tasks Scheduling in Heterogeneous Distributed Systems Population-based Metaheuristics for Tasks Scheduling in Heterogeneous Distributed Systems Flavia Zamfirache, Marc Frîncu, Daniela Zaharie Department of Computer Science, West University of Timişoara, Romania

More information

A STUDY OF TASK SCHEDULING IN MULTIPROCESSOR ENVIROMENT Ranjit Rajak 1, C.P.Katti 2, Nidhi Rajak 3

A STUDY OF TASK SCHEDULING IN MULTIPROCESSOR ENVIROMENT Ranjit Rajak 1, C.P.Katti 2, Nidhi Rajak 3 A STUDY OF TASK SCHEDULING IN MULTIPROCESSOR ENVIROMENT Ranjit Rajak 1, C.P.Katti, Nidhi Rajak 1 Department of Computer Science & Applications, Dr.H.S.Gour Central University, Sagar, India, ranjit.jnu@gmail.com

More information

LOAD BALANCING IN CLOUD COMPUTING

LOAD BALANCING IN CLOUD COMPUTING LOAD BALANCING IN CLOUD COMPUTING Neethu M.S 1 PG Student, Dept. of Computer Science and Engineering, LBSITW (India) ABSTRACT Cloud computing is emerging as a new paradigm for manipulating, configuring,

More information

Alpha Cut based Novel Selection for Genetic Algorithm

Alpha Cut based Novel Selection for Genetic Algorithm Alpha Cut based Novel for Genetic Algorithm Rakesh Kumar Professor Girdhar Gopal Research Scholar Rajesh Kumar Assistant Professor ABSTRACT Genetic algorithm (GA) has several genetic operators that can

More information

ESQUIVEL S.C., GATICA C. R., GALLARD R.H.

ESQUIVEL S.C., GATICA C. R., GALLARD R.H. 62/9,1*7+(3$5$//(/7$6.6&+('8/,1*352%/(0%

More information

Efficient Scheduling of Arbitrary Task Graphs to Multiprocessors using A Parallel Genetic Algorithm

Efficient Scheduling of Arbitrary Task Graphs to Multiprocessors using A Parallel Genetic Algorithm Efficient Scheduling of Arbitrary Task Graphs to Multiprocessors using A Parallel Genetic Algorithm Yu-Kwong Kwok and Ishfaq Ahmad Department of Computer Science The Hong Kong University of Science and

More information

Genetic Algorithms. Part 2: The Knapsack Problem. Spring 2009 Instructor: Dr. Masoud Yaghini

Genetic Algorithms. Part 2: The Knapsack Problem. Spring 2009 Instructor: Dr. Masoud Yaghini Genetic Algorithms Part 2: The Knapsack Problem Spring 2009 Instructor: Dr. Masoud Yaghini Outline Genetic Algorithms: Part 2 Problem Definition Representations Fitness Function Handling of Constraints

More information

A Locality Enhanced Scheduling Method for Multiple MapReduce Jobs In a Workflow Application

A Locality Enhanced Scheduling Method for Multiple MapReduce Jobs In a Workflow Application 2012 International Conference on Information and Computer Applications (ICICA 2012) IPCSIT vol. 24 (2012) (2012) IACSIT Press, Singapore A Locality Enhanced Scheduling Method for Multiple MapReduce Jobs

More information

A Robust Dynamic Load-balancing Scheme for Data Parallel Application on Message Passing Architecture

A Robust Dynamic Load-balancing Scheme for Data Parallel Application on Message Passing Architecture A Robust Dynamic Load-balancing Scheme for Data Parallel Application on Message Passing Architecture Yangsuk Kee Department of Computer Engineering Seoul National University Seoul, 151-742, Korea Soonhoi

More information

An ACO-based Approach for Scheduling Task Graphs with Communication Costs

An ACO-based Approach for Scheduling Task Graphs with Communication Costs An ACO-based Approach for Scheduling Task Graphs with Communication Costs Markus Bank Udo Hönig Wolfram Schiffmann FernUniversität Hagen Lehrgebiet Rechnerarchitektur 58084 Hagen, Germany {Markus.Bank,

More information

A Study of Crossover Operators for Genetic Algorithm and Proposal of a New Crossover Operator to Solve Open Shop Scheduling Problem

A Study of Crossover Operators for Genetic Algorithm and Proposal of a New Crossover Operator to Solve Open Shop Scheduling Problem American Journal of Industrial and Business Management, 2016, 6, 774-789 Published Online June 2016 in SciRes. http://www.scirp.org/journal/ajibm http://dx.doi.org/10.4236/ajibm.2016.66071 A Study of Crossover

More information

A Robust Method for Solving Transcendental Equations

A Robust Method for Solving Transcendental Equations www.ijcsi.org 413 A Robust Method for Solving Transcendental Equations Md. Golam Moazzam, Amita Chakraborty and Md. Al-Amin Bhuiyan Department of Computer Science and Engineering, Jahangirnagar University,

More information

SCHEDULING MULTIPROCESSOR TASKS WITH GENETIC ALGORITHMS

SCHEDULING MULTIPROCESSOR TASKS WITH GENETIC ALGORITHMS SCHEDULING MULTIPROCESSOR TASKS WITH GENETIC ALGORITHMS MARIN GOLUB Department of Electronics, Microelectronics, Computer and Intelligent Systems Faculty of Electrical Engineering and Computing University

More information

PROCESS OF LOAD BALANCING IN CLOUD COMPUTING USING GENETIC ALGORITHM

PROCESS OF LOAD BALANCING IN CLOUD COMPUTING USING GENETIC ALGORITHM PROCESS OF LOAD BALANCING IN CLOUD COMPUTING USING GENETIC ALGORITHM Md. Shahjahan Kabir 1, Kh. Mohaimenul Kabir 2 and Dr. Rabiul Islam 3 1 Dept. of CSE, Dhaka International University, Dhaka, Bangladesh

More information

Empirically Identifying the Best Genetic Algorithm for Covering Array Generation

Empirically Identifying the Best Genetic Algorithm for Covering Array Generation Empirically Identifying the Best Genetic Algorithm for Covering Array Generation Liang Yalan 1, Changhai Nie 1, Jonathan M. Kauffman 2, Gregory M. Kapfhammer 2, Hareton Leung 3 1 Department of Computer

More information

Skill-based Resource Allocation using Genetic Algorithms and Ontologies

Skill-based Resource Allocation using Genetic Algorithms and Ontologies Skill-based Resource Allocation using Genetic Algorithms and Ontologies Kushan Nammuni 1, John Levine 2 & John Kingston 2 1 Department of Biomedical Informatics, Eastman Institute for Oral Health Care

More information

HYBRID GENETIC ALGORITHMS FOR SCHEDULING ADVERTISEMENTS ON A WEB PAGE

HYBRID GENETIC ALGORITHMS FOR SCHEDULING ADVERTISEMENTS ON A WEB PAGE HYBRID GENETIC ALGORITHMS FOR SCHEDULING ADVERTISEMENTS ON A WEB PAGE Subodha Kumar University of Washington subodha@u.washington.edu Varghese S. Jacob University of Texas at Dallas vjacob@utdallas.edu

More information

Genetic Algorithm Performance with Different Selection Strategies in Solving TSP

Genetic Algorithm Performance with Different Selection Strategies in Solving TSP Proceedings of the World Congress on Engineering Vol II WCE, July 6-8,, London, U.K. Genetic Algorithm Performance with Different Selection Strategies in Solving TSP Noraini Mohd Razali, John Geraghty

More information

PGGA: A Predictable and Grouped Genetic Algorithm for Job Scheduling. Abstract

PGGA: A Predictable and Grouped Genetic Algorithm for Job Scheduling. Abstract PGGA: A Predictable and Grouped Genetic Algorithm for Job Scheduling Maozhen Li and Bin Yu School of Engineering and Design, Brunel University, Uxbridge, UB8 3PH, UK {Maozhen.Li, Bin.Yu}@brunel.ac.uk Man

More information

A SIMULATION MODEL FOR RESOURCE CONSTRAINED SCHEDULING OF MULTIPLE PROJECTS

A SIMULATION MODEL FOR RESOURCE CONSTRAINED SCHEDULING OF MULTIPLE PROJECTS A SIMULATION MODEL FOR RESOURCE CONSTRAINED SCHEDULING OF MULTIPLE PROJECTS B. Kanagasabapathi 1 and K. Ananthanarayanan 2 Building Technology and Construction Management Division, Department of Civil

More information

A Genetic Algorithm Approach for Solving a Flexible Job Shop Scheduling Problem

A Genetic Algorithm Approach for Solving a Flexible Job Shop Scheduling Problem A Genetic Algorithm Approach for Solving a Flexible Job Shop Scheduling Problem Sayedmohammadreza Vaghefinezhad 1, Kuan Yew Wong 2 1 Department of Manufacturing & Industrial Engineering, Faculty of Mechanical

More information

Research on a Heuristic GA-Based Decision Support System for Rice in Heilongjiang Province

Research on a Heuristic GA-Based Decision Support System for Rice in Heilongjiang Province Research on a Heuristic GA-Based Decision Support System for Rice in Heilongjiang Province Ran Cao 1,1, Yushu Yang 1, Wei Guo 1, 1 Engineering college of Northeast Agricultural University, Haerbin, China

More information

Journal of Theoretical and Applied Information Technology 20 th July 2015. Vol.77. No.2 2005-2015 JATIT & LLS. All rights reserved.

Journal of Theoretical and Applied Information Technology 20 th July 2015. Vol.77. No.2 2005-2015 JATIT & LLS. All rights reserved. EFFICIENT LOAD BALANCING USING ANT COLONY OPTIMIZATION MOHAMMAD H. NADIMI-SHAHRAKI, ELNAZ SHAFIGH FARD, FARAMARZ SAFI Department of Computer Engineering, Najafabad branch, Islamic Azad University, Najafabad,

More information

Highway Maintenance Scheduling Using Genetic Algorithm with Microscopic Traffic Simulation

Highway Maintenance Scheduling Using Genetic Algorithm with Microscopic Traffic Simulation Wang, Cheu and Fwa 1 Word Count: 6955 Highway Maintenance Scheduling Using Genetic Algorithm with Microscopic Traffic Simulation Ying Wang Research Scholar Department of Civil Engineering National University

More information

An Efficient load balancing using Genetic algorithm in Hierarchical structured distributed system

An Efficient load balancing using Genetic algorithm in Hierarchical structured distributed system An Efficient load balancing using Genetic algorithm in Hierarchical structured distributed system Priyanka Gonnade 1, Sonali Bodkhe 2 Mtech Student Dept. of CSE, Priyadarshini Instiute of Engineering and

More information

HYBRID ACO-IWD OPTIMIZATION ALGORITHM FOR MINIMIZING WEIGHTED FLOWTIME IN CLOUD-BASED PARAMETER SWEEP EXPERIMENTS

HYBRID ACO-IWD OPTIMIZATION ALGORITHM FOR MINIMIZING WEIGHTED FLOWTIME IN CLOUD-BASED PARAMETER SWEEP EXPERIMENTS HYBRID ACO-IWD OPTIMIZATION ALGORITHM FOR MINIMIZING WEIGHTED FLOWTIME IN CLOUD-BASED PARAMETER SWEEP EXPERIMENTS R. Angel Preethima 1, Margret Johnson 2 1 Student, Computer Science and Engineering, Karunya

More information

PEELSCHED: a Simple and Parallel Scheduling Algorithm for Static Taskgraphs

PEELSCHED: a Simple and Parallel Scheduling Algorithm for Static Taskgraphs PEELSCHED: a Simple and Parallel Scheduling Algorithm for Static Taskgraphs Jörg Keller, Rainer Gerhards Faculty of Mathematics and Computer Science FernUniversität in Hagen 58084 Hagen, Germany joerg.keller

More information

Path Selection Methods for Localized Quality of Service Routing

Path Selection Methods for Localized Quality of Service Routing Path Selection Methods for Localized Quality of Service Routing Xin Yuan and Arif Saifee Department of Computer Science, Florida State University, Tallahassee, FL Abstract Localized Quality of Service

More information

Multi-Objective Genetic Test Generation for Systems-on-Chip Hardware Verification

Multi-Objective Genetic Test Generation for Systems-on-Chip Hardware Verification Multi-Objective Genetic Test Generation for Systems-on-Chip Hardware Verification Adriel Cheng Cheng-Chew Lim The University of Adelaide, Australia 5005 Abstract We propose a test generation method employing

More information

Load Balancing with Task Subtraction of Same Nodes

Load Balancing with Task Subtraction of Same Nodes Load Balancing with Task Subtraction of Same Nodes Ranjan Kumar Mondal 1, Debabrata Sarddar 2 1 Research Scholar, 2 Assistant Professor 1,2 Department of Computer Science and Engineering, University of

More information

A performance study of multiprocessor task scheduling algorithms

A performance study of multiprocessor task scheduling algorithms J Supercomput (2008) 43: 77 97 DOI 10.1007/s11227-007-0139-z A performance study of multiprocessor task scheduling algorithms Shiyuan Jin Guy Schiavone Damla Turgut Published online: 1 June 2007 Springer

More information

Approximation Algorithms

Approximation Algorithms Approximation Algorithms or: How I Learned to Stop Worrying and Deal with NP-Completeness Ong Jit Sheng, Jonathan (A0073924B) March, 2012 Overview Key Results (I) General techniques: Greedy algorithms

More information

International Journal of Software and Web Sciences (IJSWS) www.iasir.net

International Journal of Software and Web Sciences (IJSWS) www.iasir.net International Association of Scientific Innovation and Research (IASIR) (An Association Unifying the Sciences, Engineering, and Applied Research) ISSN (Print): 2279-0063 ISSN (Online): 2279-0071 International

More information

Comparison of algorithms for automated university scheduling

Comparison of algorithms for automated university scheduling Comparison of algorithms for automated university scheduling Hugo Sandelius Simon Forssell Degree Project in Computer Science, DD143X Supervisor: Pawel Herman Examiner: Örjan Ekeberg CSC, KTH April 29,

More information

A Hybrid Tabu Search Method for Assembly Line Balancing

A Hybrid Tabu Search Method for Assembly Line Balancing Proceedings of the 7th WSEAS International Conference on Simulation, Modelling and Optimization, Beijing, China, September 15-17, 2007 443 A Hybrid Tabu Search Method for Assembly Line Balancing SUPAPORN

More information

GENETIC-BASED SOLUTIONS FOR INDEPENDENT BATCH SCHEDULING IN DATA GRIDS

GENETIC-BASED SOLUTIONS FOR INDEPENDENT BATCH SCHEDULING IN DATA GRIDS GENETIC-BASED SOLUTIONS FOR INDEPENDENT BATCH SCHEDULING IN DATA GRIDS Joanna Ko lodziej Cracow University of Technology, Poland Email: jokolodziej@pk.edu.pl Samee U. Khan North Dakota State University

More information

Static Load Balancing of Parallel PDE Solver for Distributed Computing Environment

Static Load Balancing of Parallel PDE Solver for Distributed Computing Environment Static Load Balancing of Parallel PDE Solver for Distributed Computing Environment Shuichi Ichikawa and Shinji Yamashita Department of Knowledge-based Information Engineering, Toyohashi University of Technology

More information

Load balancing in a heterogeneous computer system by self-organizing Kohonen network

Load balancing in a heterogeneous computer system by self-organizing Kohonen network Bull. Nov. Comp. Center, Comp. Science, 25 (2006), 69 74 c 2006 NCC Publisher Load balancing in a heterogeneous computer system by self-organizing Kohonen network Mikhail S. Tarkov, Yakov S. Bezrukov Abstract.

More information

Introduction to Scheduling Theory

Introduction to Scheduling Theory Introduction to Scheduling Theory Arnaud Legrand Laboratoire Informatique et Distribution IMAG CNRS, France arnaud.legrand@imag.fr November 8, 2004 1/ 26 Outline 1 Task graphs from outer space 2 Scheduling

More information

A SURVEY ON WORKFLOW SCHEDULING IN CLOUD USING ANT COLONY OPTIMIZATION

A SURVEY ON WORKFLOW SCHEDULING IN CLOUD USING ANT COLONY OPTIMIZATION Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 3, Issue. 2, February 2014,

More information

A Brief Study of the Nurse Scheduling Problem (NSP)

A Brief Study of the Nurse Scheduling Problem (NSP) A Brief Study of the Nurse Scheduling Problem (NSP) Lizzy Augustine, Morgan Faer, Andreas Kavountzis, Reema Patel Submitted Tuesday December 15, 2009 0. Introduction and Background Our interest in the

More information

Asexual Versus Sexual Reproduction in Genetic Algorithms 1

Asexual Versus Sexual Reproduction in Genetic Algorithms 1 Asexual Versus Sexual Reproduction in Genetic Algorithms Wendy Ann Deslauriers (wendyd@alumni.princeton.edu) Institute of Cognitive Science,Room 22, Dunton Tower Carleton University, 25 Colonel By Drive

More information

Scheduling. from CPUs to Clusters to Grids. MCSN N. Tonellotto Complements of Distributed Enabling Platforms

Scheduling. from CPUs to Clusters to Grids. MCSN N. Tonellotto Complements of Distributed Enabling Platforms Scheduling from CPUs to Clusters to Grids 1 Outline Terminology CPU Scheduling Real-time Scheduling Cluster Scheduling Grid Scheduling Cloud Scheduling 2 General Scheduling refers to allocate limited resources

More information

Genetic Algorithm Based Interconnection Network Topology Optimization Analysis

Genetic Algorithm Based Interconnection Network Topology Optimization Analysis Genetic Algorithm Based Interconnection Network Topology Optimization Analysis 1 WANG Peng, 2 Wang XueFei, 3 Wu YaMing 1,3 College of Information Engineering, Suihua University, Suihua Heilongjiang, 152061

More information

. 1/ CHAPTER- 4 SIMULATION RESULTS & DISCUSSION CHAPTER 4 SIMULATION RESULTS & DISCUSSION 4.1: ANT COLONY OPTIMIZATION BASED ON ESTIMATION OF DISTRIBUTION ACS possesses

More information

STUDY OF PROJECT SCHEDULING AND RESOURCE ALLOCATION USING ANT COLONY OPTIMIZATION 1

STUDY OF PROJECT SCHEDULING AND RESOURCE ALLOCATION USING ANT COLONY OPTIMIZATION 1 STUDY OF PROJECT SCHEDULING AND RESOURCE ALLOCATION USING ANT COLONY OPTIMIZATION 1 Prajakta Joglekar, 2 Pallavi Jaiswal, 3 Vandana Jagtap Maharashtra Institute of Technology, Pune Email: 1 somanprajakta@gmail.com,

More information

Improved Multiprocessor Task Scheduling Using Genetic Algorithms

Improved Multiprocessor Task Scheduling Using Genetic Algorithms From: Proceedings of the Twelfth International FLAIRS Conference. Copyright 999, AAAI (www.aaai.org). All rights reserved. Improved Multiprocessor Task Scheduling Using Genetic Algorithms Michael Bohler

More information

A Bi-Objective Approach for Cloud Computing Systems

A Bi-Objective Approach for Cloud Computing Systems A Bi-Objective Approach for Cloud Computing Systems N.Geethanjali 1, M.Ramya 2 Assistant Professor, Department of Computer Science, Christ The King Engineering College 1, 2 ABSTRACT: There are Various

More information

Efficient Scheduling in Cloud Networks Using Chakoos Evolutionary Algorithm

Efficient Scheduling in Cloud Networks Using Chakoos Evolutionary Algorithm International Journal of Computer Networks and Communications Security VOL., NO. 5, MAY 2015, 220 224 Available online at: www.ijcncs.org E-ISSN 208-980 (Online) / ISSN 2410-0595 (Print) Efficient Scheduling

More information

Learning in Abstract Memory Schemes for Dynamic Optimization

Learning in Abstract Memory Schemes for Dynamic Optimization Fourth International Conference on Natural Computation Learning in Abstract Memory Schemes for Dynamic Optimization Hendrik Richter HTWK Leipzig, Fachbereich Elektrotechnik und Informationstechnik, Institut

More information

A Comparison of Genotype Representations to Acquire Stock Trading Strategy Using Genetic Algorithms

A Comparison of Genotype Representations to Acquire Stock Trading Strategy Using Genetic Algorithms 2009 International Conference on Adaptive and Intelligent Systems A Comparison of Genotype Representations to Acquire Stock Trading Strategy Using Genetic Algorithms Kazuhiro Matsui Dept. of Computer Science

More information

A SIMULATION STUDY FOR DYNAMIC FLEXIBLE JOB SHOP SCHEDULING WITH SEQUENCE-DEPENDENT SETUP TIMES

A SIMULATION STUDY FOR DYNAMIC FLEXIBLE JOB SHOP SCHEDULING WITH SEQUENCE-DEPENDENT SETUP TIMES A SIMULATION STUDY FOR DYNAMIC FLEXIBLE JOB SHOP SCHEDULING WITH SEQUENCE-DEPENDENT SETUP TIMES by Zakaria Yahia Abdelrasol Abdelgawad A Thesis Submitted to the Faculty of Engineering at Cairo University

More information

Management of Software Projects with GAs

Management of Software Projects with GAs MIC05: The Sixth Metaheuristics International Conference 1152-1 Management of Software Projects with GAs Enrique Alba J. Francisco Chicano Departamento de Lenguajes y Ciencias de la Computación, Universidad

More information

Parallel Ray Tracing using MPI: A Dynamic Load-balancing Approach

Parallel Ray Tracing using MPI: A Dynamic Load-balancing Approach Parallel Ray Tracing using MPI: A Dynamic Load-balancing Approach S. M. Ashraful Kadir 1 and Tazrian Khan 2 1 Scientific Computing, Royal Institute of Technology (KTH), Stockholm, Sweden smakadir@csc.kth.se,

More information

A Fast Computational Genetic Algorithm for Economic Load Dispatch

A Fast Computational Genetic Algorithm for Economic Load Dispatch A Fast Computational Genetic Algorithm for Economic Load Dispatch M.Sailaja Kumari 1, M.Sydulu 2 Email: 1 Sailaja_matam@Yahoo.com 1, 2 Department of Electrical Engineering National Institute of Technology,

More information

Distributed Dynamic Load Balancing for Iterative-Stencil Applications

Distributed Dynamic Load Balancing for Iterative-Stencil Applications Distributed Dynamic Load Balancing for Iterative-Stencil Applications G. Dethier 1, P. Marchot 2 and P.A. de Marneffe 1 1 EECS Department, University of Liege, Belgium 2 Chemical Engineering Department,

More information

An Empirical Study and Analysis of the Dynamic Load Balancing Techniques Used in Parallel Computing Systems

An Empirical Study and Analysis of the Dynamic Load Balancing Techniques Used in Parallel Computing Systems An Empirical Study and Analysis of the Dynamic Load Balancing Techniques Used in Parallel Computing Systems Ardhendu Mandal and Subhas Chandra Pal Department of Computer Science and Application, University

More information

An Interactive Visualization Tool for the Analysis of Multi-Objective Embedded Systems Design Space Exploration

An Interactive Visualization Tool for the Analysis of Multi-Objective Embedded Systems Design Space Exploration An Interactive Visualization Tool for the Analysis of Multi-Objective Embedded Systems Design Space Exploration Toktam Taghavi, Andy D. Pimentel Computer Systems Architecture Group, Informatics Institute

More information

LOAD BALANCING TECHNIQUES

LOAD BALANCING TECHNIQUES LOAD BALANCING TECHNIQUES Two imporatnt characteristics of distributed systems are resource multiplicity and system transparency. In a distributed system we have a number of resources interconnected by

More information

Effect of Using Neural Networks in GA-Based School Timetabling

Effect of Using Neural Networks in GA-Based School Timetabling Effect of Using Neural Networks in GA-Based School Timetabling JANIS ZUTERS Department of Computer Science University of Latvia Raina bulv. 19, Riga, LV-1050 LATVIA janis.zuters@lu.lv Abstract: - The school

More information

Branch-and-Price Approach to the Vehicle Routing Problem with Time Windows

Branch-and-Price Approach to the Vehicle Routing Problem with Time Windows TECHNISCHE UNIVERSITEIT EINDHOVEN Branch-and-Price Approach to the Vehicle Routing Problem with Time Windows Lloyd A. Fasting May 2014 Supervisors: dr. M. Firat dr.ir. M.A.A. Boon J. van Twist MSc. Contents

More information

Classification - Examples

Classification - Examples Lecture 2 Scheduling 1 Classification - Examples 1 r j C max given: n jobs with processing times p 1,...,p n and release dates r 1,...,r n jobs have to be scheduled without preemption on one machine taking

More information

Applied Algorithm Design Lecture 5

Applied Algorithm Design Lecture 5 Applied Algorithm Design Lecture 5 Pietro Michiardi Eurecom Pietro Michiardi (Eurecom) Applied Algorithm Design Lecture 5 1 / 86 Approximation Algorithms Pietro Michiardi (Eurecom) Applied Algorithm Design

More information

Genetic Algorithms commonly used selection, replacement, and variation operators Fernando Lobo University of Algarve

Genetic Algorithms commonly used selection, replacement, and variation operators Fernando Lobo University of Algarve Genetic Algorithms commonly used selection, replacement, and variation operators Fernando Lobo University of Algarve Outline Selection methods Replacement methods Variation operators Selection Methods

More information

Numerical Research on Distributed Genetic Algorithm with Redundant

Numerical Research on Distributed Genetic Algorithm with Redundant Numerical Research on Distributed Genetic Algorithm with Redundant Binary Number 1 Sayori Seto, 2 Akinori Kanasugi 1,2 Graduate School of Engineering, Tokyo Denki University, Japan 10kme41@ms.dendai.ac.jp,

More information

CHAPTER 1 INTRODUCTION

CHAPTER 1 INTRODUCTION CHAPTER 1 INTRODUCTION 1.1 Background The command over cloud computing infrastructure is increasing with the growing demands of IT infrastructure during the changed business scenario of the 21 st Century.

More information

Nonlinear Model Predictive Control of Hammerstein and Wiener Models Using Genetic Algorithms

Nonlinear Model Predictive Control of Hammerstein and Wiener Models Using Genetic Algorithms Nonlinear Model Predictive Control of Hammerstein and Wiener Models Using Genetic Algorithms Al-Duwaish H. and Naeem, Wasif Electrical Engineering Department/King Fahd University of Petroleum and Minerals

More information

Towards a Load Balancing in a Three-level Cloud Computing Network

Towards a Load Balancing in a Three-level Cloud Computing Network Towards a Load Balancing in a Three-level Cloud Computing Network Shu-Ching Wang, Kuo-Qin Yan * (Corresponding author), Wen-Pin Liao and Shun-Sheng Wang Chaoyang University of Technology Taiwan, R.O.C.

More information

A Framework for Performance Analysis and Tuning in Hadoop Based Clusters

A Framework for Performance Analysis and Tuning in Hadoop Based Clusters A Framework for Performance Analysis and Tuning in Hadoop Based Clusters Garvit Bansal Anshul Gupta Utkarsh Pyne LNMIIT, Jaipur, India Email: [garvit.bansal anshul.gupta utkarsh.pyne] @lnmiit.ac.in Manish

More information

Multiobjective Multicast Routing Algorithm

Multiobjective Multicast Routing Algorithm Multiobjective Multicast Routing Algorithm Jorge Crichigno, Benjamín Barán P. O. Box 9 - National University of Asunción Asunción Paraguay. Tel/Fax: (+9-) 89 {jcrichigno, bbaran}@cnc.una.py http://www.una.py

More information

INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING & TECHNOLOGY (IJCET)

INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING & TECHNOLOGY (IJCET) INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING & TECHNOLOGY (IJCET) International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 6367(Print), ISSN 0976 6367(Print) ISSN 0976 6375(Online)

More information

Offline sorting buffers on Line

Offline sorting buffers on Line Offline sorting buffers on Line Rohit Khandekar 1 and Vinayaka Pandit 2 1 University of Waterloo, ON, Canada. email: rkhandekar@gmail.com 2 IBM India Research Lab, New Delhi. email: pvinayak@in.ibm.com

More information

Hybrid Cooperative Coevolution for Fuzzy Flexible Job Shop Scheduling Problems

Hybrid Cooperative Coevolution for Fuzzy Flexible Job Shop Scheduling Problems Hybrid Cooperative Coevolution for Fuzzy Flexible Job Shop Scheduling Problems Juan José Palacios, Camino R. Vela, Jorge Puente Dep. Computer Science University of Oviedo (Spain) {palaciosjuan,crvela,puente}@uniovi.es

More information

D A T A M I N I N G C L A S S I F I C A T I O N

D A T A M I N I N G C L A S S I F I C A T I O N D A T A M I N I N G C L A S S I F I C A T I O N FABRICIO VOZNIKA LEO NARDO VIA NA INTRODUCTION Nowadays there is huge amount of data being collected and stored in databases everywhere across the globe.

More information

A Comparison of Eleven Static Heuristics for Mapping a Class of Independent Tasks onto Heterogeneous Distributed Computing Systems 1

A Comparison of Eleven Static Heuristics for Mapping a Class of Independent Tasks onto Heterogeneous Distributed Computing Systems 1 Journal of Parallel and Distributed Computing 61, 810837 (2001) doi:10.1006jpdc.2000.1714, available online at http:www.idealibrary.com on A Comparison of Eleven Static Heuristics for Mapping a Class of

More information

On heijunka design of assembly load balancing problem: Genetic algorithm & ameliorative procedure-combined approach

On heijunka design of assembly load balancing problem: Genetic algorithm & ameliorative procedure-combined approach International Journal of Intelligent Information Systems 2015; 4(2-1): 49-58 Published online February 10, 2015 (http://www.sciencepublishinggroup.com/j/ijiis) doi: 10.11648/j.ijiis.s.2015040201.17 ISSN:

More information

Simple Population Replacement Strategies for a Steady-State Multi-Objective Evolutionary Algorithm

Simple Population Replacement Strategies for a Steady-State Multi-Objective Evolutionary Algorithm Simple Population Replacement Strategies for a Steady-State Multi-Objective Evolutionary Christine L. Mumford School of Computer Science, Cardiff University PO Box 916, Cardiff CF24 3XF, United Kingdom

More information

A probabilistic multi-tenant model for virtual machine mapping in cloud systems

A probabilistic multi-tenant model for virtual machine mapping in cloud systems A probabilistic multi-tenant model for virtual machine mapping in cloud systems Zhuoyao Wang, Majeed M. Hayat, Nasir Ghani, and Khaled B. Shaban Department of Electrical and Computer Engineering, University

More information

An optimisation framework for determination of capacity in railway networks

An optimisation framework for determination of capacity in railway networks CASPT 2015 An optimisation framework for determination of capacity in railway networks Lars Wittrup Jensen Abstract Within the railway industry, high quality estimates on railway capacity is crucial information,

More information

Real-Time Scheduling (Part 1) (Working Draft) Real-Time System Example

Real-Time Scheduling (Part 1) (Working Draft) Real-Time System Example Real-Time Scheduling (Part 1) (Working Draft) Insup Lee Department of Computer and Information Science School of Engineering and Applied Science University of Pennsylvania www.cis.upenn.edu/~lee/ CIS 41,

More information

A Comparative Performance Analysis of Load Balancing Algorithms in Distributed System using Qualitative Parameters

A Comparative Performance Analysis of Load Balancing Algorithms in Distributed System using Qualitative Parameters A Comparative Performance Analysis of Load Balancing Algorithms in Distributed System using Qualitative Parameters Abhijit A. Rajguru, S.S. Apte Abstract - A distributed system can be viewed as a collection

More information

New binary representation in Genetic Algorithms for solving TSP by mapping permutations to a list of ordered numbers

New binary representation in Genetic Algorithms for solving TSP by mapping permutations to a list of ordered numbers Proceedings of the 5th WSEAS Int Conf on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS AND CYBERNETICS, Venice, Italy, November 0-, 006 363 New binary representation in Genetic Algorithms for solving

More information

A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II

A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II 182 IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION, VOL. 6, NO. 2, APRIL 2002 A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II Kalyanmoy Deb, Associate Member, IEEE, Amrit Pratap, Sameer Agarwal,

More information

CLOUD DATABASE ROUTE SCHEDULING USING COMBANATION OF PARTICLE SWARM OPTIMIZATION AND GENETIC ALGORITHM

CLOUD DATABASE ROUTE SCHEDULING USING COMBANATION OF PARTICLE SWARM OPTIMIZATION AND GENETIC ALGORITHM CLOUD DATABASE ROUTE SCHEDULING USING COMBANATION OF PARTICLE SWARM OPTIMIZATION AND GENETIC ALGORITHM *Shabnam Ghasemi 1 and Mohammad Kalantari 2 1 Deparment of Computer Engineering, Islamic Azad University,

More information

ANT COLONY OPTIMIZATION ALGORITHM FOR RESOURCE LEVELING PROBLEM OF CONSTRUCTION PROJECT

ANT COLONY OPTIMIZATION ALGORITHM FOR RESOURCE LEVELING PROBLEM OF CONSTRUCTION PROJECT ANT COLONY OPTIMIZATION ALGORITHM FOR RESOURCE LEVELING PROBLEM OF CONSTRUCTION PROJECT Ying XIONG 1, Ya Ping KUANG 2 1. School of Economics and Management, Being Jiaotong Univ., Being, China. 2. College

More information

Scheduling Shop Scheduling. Tim Nieberg

Scheduling Shop Scheduling. Tim Nieberg Scheduling Shop Scheduling Tim Nieberg Shop models: General Introduction Remark: Consider non preemptive problems with regular objectives Notation Shop Problems: m machines, n jobs 1,..., n operations

More information

Keywords: Dynamic Load Balancing, Process Migration, Load Indices, Threshold Level, Response Time, Process Age.

Keywords: Dynamic Load Balancing, Process Migration, Load Indices, Threshold Level, Response Time, Process Age. Volume 3, Issue 10, October 2013 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Load Measurement

More information

A Binary Model on the Basis of Imperialist Competitive Algorithm in Order to Solve the Problem of Knapsack 1-0

A Binary Model on the Basis of Imperialist Competitive Algorithm in Order to Solve the Problem of Knapsack 1-0 212 International Conference on System Engineering and Modeling (ICSEM 212) IPCSIT vol. 34 (212) (212) IACSIT Press, Singapore A Binary Model on the Basis of Imperialist Competitive Algorithm in Order

More information

Expert Systems with Applications

Expert Systems with Applications Expert Systems with Applications 38 (2011) 8403 8413 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa A knowledge-based evolutionary

More information

An ant colony optimization for single-machine weighted tardiness scheduling with sequence-dependent setups

An ant colony optimization for single-machine weighted tardiness scheduling with sequence-dependent setups Proceedings of the 6th WSEAS International Conference on Simulation, Modelling and Optimization, Lisbon, Portugal, September 22-24, 2006 19 An ant colony optimization for single-machine weighted tardiness

More information

Computing Load Aware and Long-View Load Balancing for Cluster Storage Systems

Computing Load Aware and Long-View Load Balancing for Cluster Storage Systems 215 IEEE International Conference on Big Data (Big Data) Computing Load Aware and Long-View Load Balancing for Cluster Storage Systems Guoxin Liu and Haiying Shen and Haoyu Wang Department of Electrical

More information

M. Sugumaran / (IJCSIT) International Journal of Computer Science and Information Technologies, Vol. 2 (3), 2011, 1001-1006

M. Sugumaran / (IJCSIT) International Journal of Computer Science and Information Technologies, Vol. 2 (3), 2011, 1001-1006 A Design of Centralized Meeting Scheduler with Distance Metrics M. Sugumaran Department of Computer Science and Engineering,Pondicherry Engineering College, Puducherry, India. Abstract Meeting scheduling

More information

An Evolutionary Algorithm in Grid Scheduling by multiobjective Optimization using variants of NSGA

An Evolutionary Algorithm in Grid Scheduling by multiobjective Optimization using variants of NSGA International Journal of Scientific and Research Publications, Volume 2, Issue 9, September 2012 1 An Evolutionary Algorithm in Grid Scheduling by multiobjective Optimization using variants of NSGA Shahista

More information

Fair Scheduling Algorithm with Dynamic Load Balancing Using In Grid Computing

Fair Scheduling Algorithm with Dynamic Load Balancing Using In Grid Computing Research Inventy: International Journal Of Engineering And Science Vol.2, Issue 10 (April 2013), Pp 53-57 Issn(e): 2278-4721, Issn(p):2319-6483, Www.Researchinventy.Com Fair Scheduling Algorithm with Dynamic

More information