Load Balancing of Web Server System Using Service Queue Length

Transcription

1 Load Balancing of Web Server System Using Service Queue Length Brajendra Kumar 1, Dr. Vineet Richhariya 2 1 M.tech Scholar (CSE) LNCT, Bhopal 2 HOD (CSE), LNCT, Bhopal Abstract- In this paper, we describe on the load balancing system of the web server system. Web based service are used too many organization to support their customers and employee. An important issue in developing such services is ensuring the quality of service (QOS). In recent years many researcher try it to the maintained the quality of web server, with load balancer and to use various techniques. In this paper work to use on that the weighted least connection (WLC) and Round Robin (RR) algorithm comparative analysis with proposed mechanism is PC- load balancing algorithm for the load balance of the web server with service queue length. These techniques are very effective to handle the dropping rate and performance throughput or delivery of packet to view the client side. Now we consider these parameters as for as memory load for the back end to supported by the server side. Based on these parameters, we are calculating the server load and also create rules for request allocation on the web server. Keywords- web server system, load balancer, weighted least connection and round robin mechanism, service queue length with memory size. I. INTRODUCTION Load balancing to be provide a very useful to us for the web server system. Recently years to be more researchers to analyze do that effectively services to be provide for the web server system. Now here to mentioned that the web server load balancing system. The load balancing system has been found to provide an effective and scalable way of managing the ever- increasing web traffic. Although we try to get performance characteristics of the system, that user a load balancer. Typically a load balancing method or strategy is used to decide how the load balancers choose. Where to send the request there are many strategies available depending on the vender; however a few common ones are discussed this section. With that Round robin method of load balancing, the load balancer will send traffic to each node in succession. This method will evenly distributed the traffic but does not take into account the current load or responsiveness of the nodes, and weighted Least connection like the least connection methods, these load balancing method selects pool methods or nodes based on the member of active connections. However the weighted least connection methods also based their selection on server capacity. Weighted least connection method work best in environments where the servers have different capacity. This method is more intelligent, but if the connection features is enabled, the weighted least connection methods do not include the connection in the calculation when selecting a pool member or node. The weighted least connection methods uses only achieve connection in their calculation. All load balancer are capable of making traffic decision based on traditional OSI layer 2-3 information. More advance load balancer, however can make intelligent traffic management decision based on specific layers 4-7 information constrained with in the request issued by the client, such application is required in many application environments, including, those in which is a request for application data can only be met by a specific server or set of servers. Load balancing decision are made quickly, usually in less than on millisecond and high- performance load balancers can make millions of decision per second. An administrator select algorithm implemented by the load balancer determines. The physical or virtual server and send the request. Once the request is received and processed, the application servers send its response to the client via. The load balancer manages all bi-directional traffic between the client and server, it maps each application response, and users receive the proper response. 73

2 Load balancer can also configured to guaranteed that subsequent request from the same user, and part of the same session, are directed to same server as the original request, called application that must maintain state. II. RELATED WORK In this section describe on the related work on the current content with respect to server load balancing of the web server system. Rmana et al [2] in this paper the performance analysis of various distributed web server system load balancing algorithm based on different qualitative parameters, considering static and dynamic load balancing approaches are considered. The analysis indicates that these both static and dynamic types of algorithm can have advancements as well as weaknesses over each other. The main purpose of this paper is to help in the design of new algorithm in future by study of behavior and characteristic of various existing static and dynamic algorithm are more stable then dynamic algorithms but at a same time dynamic distributed algorithms are always considered better then static algorithm in distributed web server systems. Sarabjeet et al.[3] in this paper, author present the performance analysis of various load balancing algorithms based on different parameters like fault tolerance, overload rejection, stability, and cooperative, resource utilization considering two typical load balancing approaches static and dynamic. The analysis indicates that both static and dynamic types of algorithm can have advancement as well as weakness over each other. Static load balancing depends on the current situation of the system. This analysis is useful for those who to develop a new load balancing algorithm for web server. Zheng Wen, Lei Shi, Runjie Liu, Lin Qi and Lin Wei [4] according performance of load balancing scheduling polices in web server cluster system is greatly impacted by the characteristics of work load, based on the analysis of the load characteristics for scheduling algorithm, prediction-based adaptive load balancing model (RR_MMMCS_A_P) is proposed in this paper, the arrival rate the size of the fallow-up, request are predictive by RR_MMMCS_A_P and rapid adjustment of the corresponding parameter to balance the load between servers. Experiment have shown that compared with CPU based and CPU scheduling strategy, RR_MMM_CS_A_P have better performance in reducing average response time for both calculation-intensive and data_-intensive jobs. Cardellini et al. [6] analyze various dispatching policies under realistic situation where state information needs to be estimated. They also that packet rewriting by the dispatcher presents problems because the dispatcher must rewrite incoming as well as outgoing packets, and outgoing packets typically out number incoming request packets. Pao and Chen [7] presented the dispatcher-based load balancing architecture. This architecture is based on the decision-making based on the behavior of the server system. When the backend servers have different computation power, the system must use the real time loading information as the decision-making criteria. They proposed a dispatcher architecture that uses the remaining capacity of each backend servers to decide the most appropriate server. in addition to improve the performance, the system can also use the remaining capacity algorithm to reduce the cost project work. We have introduction a new performance parameter, queue size for selected the best appropriate web server to transfer new request. We are trying to get the reduce load of the web server system regarding above mention problem of the web server load balancing and web server architecture. We consider another performance parameter of the web server that is memory load, number of active connection and queue length of the web server. Based on these parameters we are calculating the server load and also create some rules for request allocation on the web server. These rules help us to find the best web server to transfer the new requests. III. PROPOSED METHOD In this section we describe on the proposed methodology covered with that web server load balancing technique and algorithms. 3.1 Web Server Load Balancing Algorithm Load balancing of the web server system has been the area of research in the few decades. Load balancing algorithms always try to balance the load by transferring the load from heavily loaded server to lightly loaded servers. It is the process to remaining the load to the individual server to make resource utilization effective and better response time. The main aspects to be considering when we are designing a load balancing algorithm are estimation of load, memory load, number of connection, behavior of the system. Depending on the current state of the system load balancing algorithm can be divided in to categories static and dynamic load balancing algorithms. Such as the illustrate that and discuss below a different type of the web server load balancing technique and algorithm as for as the review with the study there. 74

3 Static load balancing algorithm Dynamic load balancing algorithm Static Load Balancing Algorithm Static load balancing is [2] the simplest form of the two types. Static load balancing is the selection and placement of a logical process on some processing element located on a distributed network. The selection of the processing element for some for some logical process is based upon some weighting factor for that process, or some kind of workload characterization of the processing elements or the network topology, possibly both. The process of selecting a processing element for a logical process requesting service may be done with two possible methods, a stateless method or a state based method. With a stateless method the selection of a processing element without regards any knowledge of the state. A state-based method of the selecting a processing element of the implies that the selection of a processing element requires knowledge of the system state, either globally, or locally. Globally knowledge implies that the state of the all components of the system is known and local knowledge implies that only partially knowledge is known. If the state of the system is needed then some kind of messaging or probing of network resource among the requesting processes, agent or processing elements is needed to determine their availability. Once a processing element is selected for a logical process, the logical process is executed on the selected processing element for a logical process lifetime. Two examples of stateless placement techniques for selecting a processing element for logical process are round robin and random placement. Round robin placement selects the next processing element from a predefined list. Random placement selects an element randomly from a set of processing element. Static load balancing algorithms are of the two type round robin load balancing algorithm. Weighted Round Robin load balancing algorithm and Random Allocation algorithm Round Robin Load Balancing Algorithm Round Robin (RR) [2] algorithm distributed client requests evenly to all servers in round robin manner, meaning that server choosing is performed in series and will be back to the first server if the server has been reached. Servers choosing are performed locally on each server, independent of allocations of other server. Advantage of round robin algorithm is that it is simplest in nature. Round robin load balancing provides better performance for homogeneous environment where the entire server has same capacity. In generally cluster of the web server have different capacity. 75 RR load balancing treats the entire server equally and provides request in series, so the higher capacity server works efficiently, but low capacity server does not provide better result for same number requests Weighted Round Robin Algorithm In weighted Round Robin algorithm [2], this algorithm maintains a list weight and assigns a weight to the servers present in a cluster. These weight works as to the server. Higher weighted server has high priority and forwards new request in proportion to the weight of each server. This algorithm uses more computation times then Round Robin algorithm. However, the additional computation results in distributing the traffic more efficiently to the server that is most capable of handling the request Random Allocation Algorithm In a Random allocation, the HTTP requests are assigned to any server picked randomly among the cluster of servers. In such a case, one of the servers may be assigned many more requests to process, while the other servers are sitting idle. However, on average, each server gets its share of the load due to the random selection. This algorithm is easy to implement. This algorithm can lead to overloading of one server while other server is idle Dynamic Load Balancing Algorithm If differs from static algorithms in the workload is distributed among the servers at runtime. The load balancer assigns new requests to the server based on the information collected. In a distributed system, dynamic load balancing can be done in two different ways: distributed and nondistributed [3]. In the distributed one, the dynamic algorithm is executed by all servers present in the cluster and the task of load balancing is shared among them. In non-distributed type, either one server or clusters of servers do the task of load balancing. Non-distribution dynamic load balancing algorithm can take two forms: centralized and semi-distributed. In the first form, the load balancing algorithm is executed only by a single server in the whole system: the central server. This server is solely is responsible for load balancing of the whole cluster. The other servers interact only with the central server. In semidistributed form, servers of the large cluster are portioned into cluster by appropriate selection technique, which takes care of load balancing within that cluster [3]. Hence the load balancing of the whole system is done via the central servers of each cluster. Centralized dynamic load server decreases drastically as compared to the semi-distributed case.

4 However, centralized algorithm can render useless once the central server crashes. Therefore, this algorithm is most suitable for network with small size. In this paper several related research about load balancing algorithm and web server architecture have been discussed. We have also discussed about dispatcher-based approach, admission control mechanism of a web server and discuss distributed computing and client-server model architecture Distributed Computing In general, distributed computing is any computing that involves multiple computers remote from each that has a role in a computation problem or information processing. In business enterprises, distributed computing generally has meant putting various steps in business processes at the most efficient place in a network of computers. In the typical transaction using the 3- tire model, user interface processing is done database access and processing is done in another computer that provides centralized access for many business processes. Typically, this rather distributed computing uses the client/server communication model. Figure: - 1. Simple Distributed Architecture 3. 3 Client Server Architecture The client-server model is a distributed computing structure that partition takes or workloads between the provide of a resource or service, called server and service requesters, called clients. A client is a single-user workstation that provides presentation services and the appropriate computing, connectivity and the database services and the interface relevant to the memory providing computing, connectivity and the database services and the satisfies the business need by appropriately allocating the application processing between the client and the server processors. The protocol is the client requests the services from the server; the server processes the request and requests the result to the client. The communication mechanism is a message passing inters process communication (IPC) that enables distributed placement of the client and server processes. Figure: - 2. Simple Client-Server Architecture 3. 4 Dispatcher Based Approach To centralized request scheduling and fully control the client request routing. A dispatcher-based approach has been proposed in the paper. The dispatcher uniquely identifies each server in the system through a private address that can be a different protocol levels, depending on the architecture. Dispatching of request can be done using various technique Round Robin, lead connection etc Admission Control Mechanism Admission control mechanism is used to protected web server from overloaded and provide performance guarantee or differentiated service class of requests. Kihl et at [8] have presented a control theoretic model of the web server. They use non linear control theory. Their model is based on the control theory. They developed a non-linear stochastic control theoretic model of GI/G1 system. They also show that their model can be used to design admission control mechanism that behaves well in the real system that is the queuing system. They also develop two type of controller PI and RST controller both commonly used in automatic control. IV. PROPOSED WEB SERVER ARCHITECTURE In this section, the proposed web server architecture has analyzed based on load balancing algorithm. As we know that most of the load balancing algorithm focused on the homogeneous environment, where all web server have specification. The web servers have same processing, memory, and I/O speed etc. 76

5 In the homogeneous environment a load balancer, redirect the client request to the most appropriate Web server based on the some well-known strategies such as Round Robin and Least Connection. In a server state based load balancing, server must report their load condition to the Web server. The load balancer then selects the best server to serve the request. However, in the heterogeneous environment load balancer does not depend on the loading condition of the Web server. It is also need another parameter to redirect the request to the best Web server. In our proposed load balancing algorithm, we have considered other performance parameters such as memory load, number of active connection and queue length. We have considered performance parameter to redirect the requests and then we compared our algorithm with Round Robin algorithm and Least Connection load balancing algorithm. Round Robin (RR) [7] load balancing algorithm evenly distributes the load in the cluster of server in a circular way means it starts with first server and reaches the last server. If last servers found than starts with first server. RR load balancing algorithm gives better results in homogeneous environment where all the server have same capacity, but in the heterogeneous environment it cannot provide better results. Least Connection (LC) [7] algorithm redirects the requests to the server, which have minimum number of connection. It does not consider the current load condition of the server. If Web server is overloaded but the number of active connection is lowest, then it redirects the new request to the server. But server is overloaded so it may be drop new request. In our proposed algorithm, we have considered three parameters (number of connection, memory load and queue length are the parameter to server selection) of the server to choose the best Web server in the cluster. A simple load balancing architecture has presented in the Figure- 3. Considering a load balancing system with N number of server and one load balancer. When a user sends a request to access particular page then it sends a request to the DNS server for the IP address of the Web server, if the client does not have IP address the cache. DNS server will reply with the IP address of the load balancer instead of the Web server provides the particular page. Then client sends an HTTP request to the load balancer. When the load balancer receives the request, a redirection page will be sent back to the client. The redirection page contains the IP address of the Web server, which is the best one to serve this request. This redirection depends on the load balancing algorithm used by the load balancer. Then client issues again HTTP request to the real Web server. The Web server then will serve the client request and transfer the desire page. Figure:-3. Proposed Web Server Load Balancing Architecture We have introduced a new module in to the Web server. This module called status monitor. This status monitor has monitored the performance of the Web server periodically. Status monitor constantly sends back the status of the Web server to the load balancer. Load balancer has a load collection module, which collects current load information of each Web server and periodically calculated the actual load of the Web server. Status monitor sends current value of remaining memory, active number of connection and queue length of the Web server queue. Constantly send status report to the load balancer. The load balancer uses this status information to choose the best appropriate Web server Load Balancing Algorithm For our proposed load balancing algorithm, we are using load balancing architecture of the Web server. The functionality of this architecture we have already discussed above. We are taking three performance parameters for our proposed load balancing algorithm. Remaining memory, number of active connection and queue length is the parameters. According to these parameter load balancer find best Web server to serve request. Let the status monitor of the server, reports the current load information parameters as {x, y, z}, where x is the remaining memory, y is the number of remaining connection and z is the remaining queue length of the server. Load balancer calculates the server capacity according to these parameters. PC- load balancing mechanism of the proposed load balancing algorithm below- 77

6 Pc- Load Balancing Algorithm 1. If Q(n) (1<=n<=N) is the current queue length at server n, the expected queue length after a time interval T, [equal to the expected time between two successive] can be approximated by- Where, L (n) expected number of request assigned to server n, T time interval, is the service rate, is the mean of new arrival request in this time interval. 2., is expected number of request processing T assuming that the server is always busy. 3. (a) Initial stage: Initialized parameter of server side (e) Type[x, y, z], where x is memory capacity, y is the connection capacity, z is the queue capacity. (b) Check balance equality weighted factor with equality finder mathematically equation [X+Y+ Z=1] Where, calculate the load balancing with serving capacity (U) of server n. (c) Find the with available number of n server. V. EXPERIMENTAL ANALYSIS In this section, experiment results have been analyzed. We have compared our proposed load balancing algorithm with other two existing load balancing algorithm (Round Robin and weighted Least Connection). We have taken homogeneous as well as heterogeneous environment. 5.1 Homogeneous Environment With Fixed Queue-Size In the homogeneous environment, all the Web servers are having same configuration. For experiment, we have taken memory 1 GB, number of connections 1000 and queue size is 50 on each Web server, Table-1 shows the total generated traffic and dropped requests by Round Robin (RR), weighted Least Connection (WLC) and proposes load balancing algorithm. As we see that, drop rate of weighted Least Connection algorithm is high. Proposed load balancing algorithm shows best results in homogeneous environment with fixed queue size. (d) In server TABLE:-1 Experiment Results (RR vs. WLC vs. PC- load Algorithm) Requests Generated Requests Dropped RR WLC PC-load Algorithm

7 Figure: - 4 show the comparison of all the three algorithms. In all the cases, drop rate of proposed PC-load balancing algorithm is minimum. The server throughput comparisons of different load balancing algorithms are shown in the Figure:-5. If drop rate of any Web server is less then it serves more requests, so it will provide more throughputs comparatively. Proposed PC-load balancing algorithm has less drop rate so it will provide higher through as shown in the Figure. Figure:-4 Comparison of all three Load Balancing Algorithms homogeneous environment with fixed queue size. As we already have discussed if the drop of any server is high than the throughput of the server is low. In Round Robin and weighted Least Connection algorithm has high drop rate. Therefore, server throughput is minimum. Our Proposed algorithm has low drop rate so it provides better throughput. The formula for the server throughput given by- Server Throughput = (Total Requests-Drop Requests)/ Total Requests. Table:-2 show the throughput of the server in homogeneous environment with fixed queue size. When number of generated is 4000 requests, throughput of the Web server in proposed load balancing algorithm is more than 80%. In Round Robin and weighted Least Connection, algorithm server throughput is almost same 70%. TABLE:-2 Server Throughput in Homogeneous Environment with fixed Queuesize Requested Generated Server Throughput (in %) RR WLC Proposed Figure:-5 Server Throughput in Homogeneous environment (fixed Queue Size) 5.2 Heterogeneous Environment With Fixed Queue Size We have compared our proposed PC-Load balancing algorithm in heterogeneous environment with fixed queue size. In heterogeneous environment server have different configuration. We have taken different parameter for experiment as given in Table:-3. Table:-3 Heterogeneous Server Configuration with fixed Queue Size List of Server Memory connection Queue Size( request) Server1 1GB Server2 2GB Server3 1GB Table:- 4 shows that in the heterogeneous environment, we have generated 5000, 6000, 7000, 8000 requests for experimental analysis and we have observed that proposed load balancing gives better results than the other two load balancing algorithm (RR and WLC). 79

8 Weighted Least Connection gives worst results because it depends on the connections Server1 and Server 3 have minimum configuration 1000 parallel connection, so, according to their nature it will sends new requests to the Server1 or Server3. When servers reach their maximum serving capacity (1000 connection), after this new incoming request may dropped. Table: - 4 Experiment Result in Heterogeneous Environment with fixed Queue size Requests Generated Requests Dropped RR WLC PC-load Algorithm Table-14 shows the throughput of the servers. PC-Load balancing algorithm have high throughput of the server followed by Round Robin Algorithm and Weighted Least Connection shows worst results because it has the high drop rate. We have calculated server throughput using the formula given below: Server Throughput = (Total Requests- Drop Requests)/ Total Requests. Requested Generated Table:-5 Server Throughput with fixed Queue size Server Throughput (in %) RR WLC PC-Load Balancing Algorithm Figure: - 6 shows, Round Robin gives better results than Weighted Least Connection Algorithm, because it sends the requests to all three servers in a circular way. Weighted Least Connection has worst results than Round Robin and PC-Load Balancing Algorithm because it chooses the server having fewer requests. It sends request to server1 or Server3, server1 and Server3 have 1000 parallel connections so, it can server 1000 connections, after 1000 request it may drop the requests Figure:-7 show throughput comparison of the servers. When the incoming requests is 5000, server throughput in the proposed algorithm is more than 88.40%, Round Robin has more than 74.58% throughput. Weighted Least Connection has minimum throughput about to 61.76%. Figure: - 6. Comparison of Load Balancing Algorithm (fixed Queue size) Figure: - 7. Comparison of Throughput of the Servers (fixed Queue Size) Proposed PC- load balancing algorithm has shown higher throughput because drop rate is minimum. Round Robin has shown lesser throughput than proposed algorithm and higher throughput as compare to the weighted Least Connection algorithm. 80

9 Weighted Least Connection has shown worst throughput because the drop rate is high. In this way, we have compared proposed Load Balancing algorithm in homogeneous as well as heterogeneous environments. In both the cases, we have taken several different scenarios. In all scenarios, proposed algorithm has shown less drop rate as compared to the existing load balancing algorithms (Round Robin and weighted Least Connection). Since the drop rate of the proposed Algorithm is low, the servers have high throughput. Therefore, we can state that the proposed Load Balancing algorithm provides a better option for homogeneous as well as heterogeneous environment to handle the overloaded condition of the Web servers. VI. CONCLUSION Load balancing algorithm on the Web server system has been used to improve the availability and reduce the overloading of the Web Servers. After the comparative analysis of the various classical approaches of the load balancing, we have developed an efficient architecture of the Web server system. We have designed an efficient Load Balancing algorithm to manage the load of the Web servers using some new modules as status monitor and load collection to calculate the current loading condition of the Web servers. We have calculated the performance parameters such as remaining memory, remaining connection, and remaining queue size to calculate the current serving capacity of the Web server. We have transferred new requests to the Web servers which have maximum remaining serving capacity. Further, we have performed comparative analysis of proposed algorithm with the existing algorithms (Round Robin and weighted Least Connection). In experimental analysis, we have observed that the proposed algorithm gives better result in homogeneous as well as heterogeneous environment considering fixed as well as varying queue size. Drop rate of the proposed Load Balancing algorithm has been also observed that it is less as compared to existing Load balancing algorithms. When the drop rate of any server is less, then it gives higher throughput and Web server processes more requests. Thus, proposed Load balancing algorithm has given high server throughput and minimum drop rate of the requests even in the overloaded conditions of the Web servers. VII. FUTURE WORK Load balancing is a concept, which is still under research. Everyday new algorithms are being developed and existing models are studied. So, in the future, we can analyze the proposed Load Balancing algorithm in real time environments and calculate the serving capacity of the algorithm to achieve the throughput and utilization. Further, we can also calculate the response time of the Web servers to check its functionality. REFERENCES [1] Mikael Aderson, Introduction to web server modeling and control Research, Technical report, Octuber 28, [2] K. Ramana, A. Subramanyam and A. Ananda Rao, Comparative Analysis of distributed web server system load balancing algorithm using Qualitative parameters, VSRD International Journal of Computer Science and Information Technology, Vol. 1(8), 2011, pp [3] Sndeep Sharma, Sarabjit Singh,and Meenakshi Sharma, Performance analysis of load balancing algorithm, World Academy of science, Engineering and Technology, vol. 38, [4] Zheng Wen, Lei Shi, Lin Qi and Lin Wei, A Predictive Adaptive load Balancing Model, In proceeding of the 9 th IEEE International Conference on fuzzy and knowledge discovery (FSKD 2012), Cihna. [5] O. Damani, P. Chung, and C. Kintala, One IP: Technique for hosting a service on a cluster of machines, In proceeding of 41 st IEEE computing society International Conference, pp.85-92, April [6] Cardellini, V. Colajanni, M., and Yu, Dynamic load balancing in geographically distributed heterogeneous web server, in proceeding of the IEEE 18 th International Conference on distributed computing system, Amsterdam, the Netherlands pp , [7] T.L.Pao, J.B.Chen, the Scalability of heterogeneous Dispatcher based web server load balancing architecture, proceeding of the 7 th International Conference on parallel and distributed computing, application and technology, pp , [8] M. Kihl, A. Robbertsson, Performance modeling and control of server system using non- linear control theory, in proceeding of 18 th International Tele traffic congress, 2003 [9] M. Anderson, M. Kihl and A. Robertson, Modeling and Design of Admission Control Mechanism for Web Server Using Non-linear Control Theory In proceeding of ITCOM 03, September- 2003, Orlando, USA. [10] K. Li., S. Jamin, A Measurement Based Admission Controlled Web Server, In IEEE Infocom, Telaviv, Israel [11] Ivan Voras and Mario Zagar, Characteristics of multithreading model for high-performance IO driven network applications, The Computing Repository, Sep [12] L.K. Singh and R. Srivastava, Memory Estimation of Internet Server Using Queuing Theory: Comparative Study between M/G/1, G/M/1 & G/G/1 Queuing Model, World Academy of Science, Engineering and Technology, vol. 9,