LaPIe: Collective Communications adapted to Grid Environments
|
|
|
- Sherilyn Hodge
- 8 years ago
- Views:
Transcription
1 LaPIe: Collective Communications adapted to Grid Environments Luiz Angelo Barchet-Estefanel Thesis Supervisor: M Denis TRYSTRAM Co-Supervisor: M Grégory MOUNIE ID-IMAG Laboratory Grenoble - France LaPIe: Collective Communications adapted to Grid Environments 1 / 60
2 Introduction to Parallel Processing Fact The demand for computing power will always grow up There are two options to increase the available computer power: LaPIe: Collective Communications adapted to Grid Environments 2 / 60
3 Introduction to Parallel Processing Fact The demand for computing power will always grow up There are two options to increase the available computer power: Buy a bigger computer - $$$$$ LaPIe: Collective Communications adapted to Grid Environments 2 / 60
4 Introduction to Parallel Processing Fact The demand for computing power will always grow up There are two options to increase the available computer power: Buy a bigger computer - $$$$$ Use several computers Parallel Processing Divide a problem into multiple fragments that can be executed in parallel LaPIe: Collective Communications adapted to Grid Environments 2 / 60
5 Introduction to Grids/Metacomputing Definition Aggregation of geographically distributed computers Mainly clusters of computers Fact The Grid hardware already exists Interconnexion of several clusters and NOWs The Grid software only emerges Most difficulties come from the resource heterogeneity LaPIe: Collective Communications adapted to Grid Environments 3 / 60
6 Communications in a Grid Influence of resource heterogeneity Geographically distributed systems Different communication latencies Heterogeneous communication infrastructures Transfer bandwidth LaPIe: Collective Communications adapted to Grid Environments 4 / 60
7 Example: GRID 5000 Latency Bandwidth* Myrinet 10 µs 250 MB/s Giga Ethernet 50 µs 120 MB/s WAN Connection 5000 µs MB/s average bandwidth for a 32MB message sent with MPI LaPIe: Collective Communications adapted to Grid Environments 5 / 60
8 Example: GRID 5000 Latency Bandwidth Myrinet 10 µs 250 MB/s Giga Ethernet 50 µs 120 MB/s WAN Connection 5000 µs MB/s average bandwidth for a 32MB message sent with MPI LaPIe: Collective Communications adapted to Grid Environments 6 / 60
9 Collective Communications Definition Collective communication is defined as communication that involves a group of processes Different communication patterns LaPIe: Collective Communications adapted to Grid Environments 7 / 60
10 Collective Communications Definition Collective communication is defined as communication that involves a group of processes Different communication patterns Most programming environments include collective communication primitives PVM, MPI, Athapascan, etc. Consensus, Group Membership, etc. LaPIe: Collective Communications adapted to Grid Environments 7 / 60
11 Collective Communications Impact of communication heterogeneity Absence of a single efficient strategy LaPIe: Collective Communications adapted to Grid Environments 8 / 60
12 Collective Communications Impact of communication heterogeneity Absence of a single efficient strategy Performance depends on: communication pattern network characteristics operation parameters (# of nodes, message size, etc.) LaPIe: Collective Communications adapted to Grid Environments 8 / 60
13 Overview of this work Our goal: improve communication scheduling on grid environments through the use of an hierarchical network modelling provide efficient grid-aware collective communication operations What we need: qualitative knowledge of the network topology detect network heterogeneity quantitative knowledge of the network interconnexions identify latency and bandwidth among different nodes LaPIe: Collective Communications adapted to Grid Environments 9 / 60
14 Overview of this work Our approach: use hybrid algorithms dynamic scheduling of inter-cluster communication efficient static algorithms for intra-cluster communication Technical validation: evaluation through synthetic experiences performances are close to those experienced by real applications fast prototyping LaPIe: Collective Communications adapted to Grid Environments 10 / 60
15 Outline 1 Optimising Collective Communications LaPIe: Collective Communications adapted to Grid Environments 11 / 60
16 Optimising Collective Communications Objective: minimise the overall execution time improve data distribution reduce communications through slow links LaPIe: Collective Communications adapted to Grid Environments 12 / 60
17 Optimising Collective Communications Objective: minimise the overall execution time improve data distribution reduce communications through slow links Heterogeneous Systems - Grids communication scheduling according to the network characteristics LaPIe: Collective Communications adapted to Grid Environments 12 / 60
18 Optimising Collective Communications Objective: minimise the overall execution time improve data distribution reduce communications through slow links Heterogeneous Systems - Grids communication scheduling according to the network characteristics NP-Complete no accurate analytical models are available LaPIe: Collective Communications adapted to Grid Environments 12 / 60
19 Hierarchical Structure Flat Tree approach LaPIe: Collective Communications adapted to Grid Environments 13 / 60
20 Hierarchical Structure Flat Tree approach Objective: minimise distant communications LaPIe: Collective Communications adapted to Grid Environments 13 / 60
21 Hierarchical Structure Flat Tree approach Objective: minimise distant communications Communication is divided in two layers LaPIe: Collective Communications adapted to Grid Environments 13 / 60
22 Hierarchical Structure Flat Tree approach Objective: minimise distant communications Communication is divided in two layers Distant nodes LaPIe: Collective Communications adapted to Grid Environments 13 / 60
23 Hierarchical Structure Flat Tree approach Objective: minimise distant communications Communication is divided in two layers Distant nodes Local nodes LaPIe: Collective Communications adapted to Grid Environments 13 / 60
24 Hierarchical Structure Flat Tree approach Objective: minimise distant communications Communication is divided in two layers Distant nodes Local nodes ECO (Lowekamp 96) - PVM library MagPIe (Kielmann 99) - MPI library LaPIe: Collective Communications adapted to Grid Environments 13 / 60
25 Analysis of this approach Advantages Easy to implement Minimises communication across slow links Limitations Too tight scheduling communication hierarchy does not make difference between links capacities/latencies The root process handles all long distance transmissions does not explore parallel transmissions LaPIe: Collective Communications adapted to Grid Environments 14 / 60
26 Multi-layered Hierarchy Multi-layered communications LaPIe: Collective Communications adapted to Grid Environments 15 / 60
27 Multi-layered Hierarchy Multi-layered communications Structured according to the relative performance of each layer WAN > MAN > LAN > SMP LaPIe: Collective Communications adapted to Grid Environments 15 / 60
28 Multi-layered Hierarchy Multi-layered communications Structured according to the relative performance of each layer WAN > MAN > LAN > SMP MPICH-G2 (Karonis 02) - MPI library LaPIe: Collective Communications adapted to Grid Environments 15 / 60
29 Analysis of this approach Advantages More flexible structure Based on the relative communication performance Limitation Hierarchy does not takes into account the communication cost inside each cluster LaPIe: Collective Communications adapted to Grid Environments 16 / 60
30 Analysis of this approach Advantages More flexible structure Based on the relative communication performance Limitation Hierarchy does not takes into account the communication cost inside each cluster LaPIe: Collective Communications adapted to Grid Environments 16 / 60
31 How to improve Grid communications Is it possible to better schedule communications in a grid environment? Dynamically generated hierarchy network parameters, message size and communication pattern Fully Grid-aware includes the communication cost inside each cluster LaPIe: Collective Communications adapted to Grid Environments 17 / 60
32 Our Approach Simplify the network description focus on topology discovery and clustering Augment the information about clusters performance performance models to predict the communication cost Improve the usage of multi-layered hierarchy grid-aware scheduling heuristics LaPIe: Collective Communications adapted to Grid Environments 18 / 60
33 Outline 1 Optimising Collective Communications LaPIe: Collective Communications adapted to Grid Environments 19 / 60
34 Approaches Locality of the nodes User-defined mappings Network discovery tools LaPIe: Collective Communications adapted to Grid Environments 20 / 60
35 Approaches Locality of the nodes User-defined mappings Network discovery tools Locality of the nodes Simple Does not express clusters internal heterogeneity Does not consider interconnection parameters LaPIe: Collective Communications adapted to Grid Environments 20 / 60
36 Approaches Locality of the nodes User-defined mappings Network discovery tools User defined topology Expensive and hard to do Sufficiently accurate (?) Normally falls back to the locality of the nodes LaPIe: Collective Communications adapted to Grid Environments 20 / 60
37 Approaches Locality of the nodes User-defined mappings Network discovery tools Some network tools NWS - measures latency and bandwidth between nodes REMOS - uses SNMP to construct a low-level topology TopoMon - identifies shared links LaPIe: Collective Communications adapted to Grid Environments 20 / 60
38 What we need Application-level topology discovery identification of homogeneous islands fast deployment LaPIe: Collective Communications adapted to Grid Environments 21 / 60
39 Topology Discovery First Phase: identify network heterogeneity use of NWS-like tools LaPIe: Collective Communications adapted to Grid Environments 22 / 60
40 Topology Discovery First Phase: identify network heterogeneity use of NWS-like tools construct a n n distance matrix latency LaPIe: Collective Communications adapted to Grid Environments 22 / 60
41 Details How to minimise the probing time Latency measure is short enough to not disturb the network Schedule parallel probes among independent pairs LaPIe: Collective Communications adapted to Grid Environments 23 / 60
42 Topology Discovery Second Phase: clustering use of a clustering algorithm (ECO) Tolerance factor ρ = 30% LaPIe: Collective Communications adapted to Grid Environments 24 / 60
43 Topology Discovery Second Phase: clustering use of a clustering algorithm (ECO) Tolerance factor ρ = 30% Formatted output (magpie_clusters file) LaPIe: Collective Communications adapted to Grid Environments 24 / 60
44 Topology Description cluster 0 process cluster 1 process cluster 2 process cluster 3 process 24 cluster 4 process 27 cluster 5 process LaPIe: Collective Communications adapted to Grid Environments 25 / 60
45 Topology Discovery Third Phase: obtaining network parameters Reduced set of measures one node from each cluster LaPIe: Collective Communications adapted to Grid Environments 26 / 60
46 Topology Discovery Third Phase: obtaining network parameters Reduced set of measures one node from each cluster O(C 2 ) measures LaPIe: Collective Communications adapted to Grid Environments 26 / 60
47 Topology Discovery Third Phase: obtaining network parameters Reduced set of measures one node from each cluster O(C 2 ) measures Merge of this information with network topology LaPIe: Collective Communications adapted to Grid Environments 26 / 60
48 Example: the IDPOT cluster LaPIe: Collective Communications adapted to Grid Environments 27 / 60
49 Outline Optimising Collective Communications Cost model Broadcast Validating the models 1 Optimising Collective Communications LaPIe: Collective Communications adapted to Grid Environments 28 / 60
50 Cost model Broadcast Validating the models Modelling Collective Communications We use plogp cost model (Kielmann et al.) Number of processes - P Latency - L Communication gap - g(m) Send and receive overhead - os(m), or(m) g(m) os(m) g(m) g(m) L or(m) L LaPIe: Collective Communications adapted to Grid Environments 29 / 60
51 Advantages of plogp Cost model Broadcast Validating the models gap Measured Gap between GdX and IDPOT clusters microseconds Message size (bytes) LaPIe: Collective Communications adapted to Grid Environments 30 / 60
52 Cost model Broadcast Validating the models Comparing with the Hockney model gap Measured Gap between GdX and IDPOT clusters microseconds Message size (bytes) plogp allows a theoretical modelling that is close to the reality LaPIe: Collective Communications adapted to Grid Environments 31 / 60
53 Example: modelling MPI_Bcast Cost model Broadcast Validating the models Definition One process (root) send the same message to every process in the group LaPIe: Collective Communications adapted to Grid Environments 32 / 60
54 Example: modelling MPI_Bcast Cost model Broadcast Validating the models Definition One process (root) send the same message to every process in the group Strategies Flat Tree Binary Tree Binomial Tree Chain (pipeline)... LaPIe: Collective Communications adapted to Grid Environments 32 / 60
55 Example: modelling MPI_Bcast Cost model Broadcast Validating the models Definition One process (root) send the same message to every process in the group Strategies Flat Tree Binary Tree Binomial Tree Chain (pipeline)... LaPIe: Collective Communications adapted to Grid Environments 32 / 60
56 Example: modelling MPI_Bcast Cost model Broadcast Validating the models Definition One process (root) send the same message to every process in the group Strategies Flat Tree Binary Tree Binomial Tree Chain (pipeline)... LaPIe: Collective Communications adapted to Grid Environments 32 / 60
57 Example: modelling MPI_Bcast Cost model Broadcast Validating the models Definition One process (root) send the same message to every process in the group Strategies Flat Tree Binary Tree Binomial Tree Chain (pipeline)... LaPIe: Collective Communications adapted to Grid Environments 32 / 60
58 Cost model Broadcast Validating the models MPI_Bcast Modelling on Homogeneous Clusters Implementation Strategy Communication Model Flat Tree (P 1) g(m) + L Flat Tree with Rendez-vous (P 1) g(m) + 2 g(1) + 3 L Segmented Flat Tree (P 1) (g(s) k) + L Binomial Tree log 2 P g(m) + log 2 P L Binomial Tree with Rendez-vous log 2 P g(m) + log 2 P (2 g(1) + 3 L) Segmented Binomial Tree log 2 P g(s) k + log 2 P L Binary Tree log 2 P (2 g(m) + L) Chain (P 1) (g(m) + L) Chain with Rendez-vous (P 1) (g(m) + 2 g(1) + 3 L) Segmented Chain (Pipeline) (P 1) (g(s) + L) + (g(s) (k 1)) LaPIe: Collective Communications adapted to Grid Environments 33 / 60
59 Cost model Broadcast Validating the models MPI_Bcast Modelling on Homogeneous Clusters Implementation Strategy Communication Model Flat Tree (P 1) g(m) + L Flat Tree with Rendez-vous (P 1) g(m) + 2 g(1) + 3 L Segmented Flat Tree (P 1) (g(s) k) + L Binomial Tree log 2 P g(m) + log 2 P L Binomial Tree with Rendez-vous log 2 P g(m) + log 2 P (2 g(1) + 3 L) Segmented Binomial Tree log 2 P g(s) k + log 2 P L Binary Tree log 2 P (2 g(m) + L) Chain (P 1) (g(m) + L) Chain with Rendez-vous (P 1) (g(m) + 2 g(1) + 3 L) Segmented Chain (Pipeline) (P 1) (g(s) + L) + (g(s) (k 1)) LaPIe: Collective Communications adapted to Grid Environments 33 / 60
60 Flat Tree Broadcast Cost model Broadcast Validating the models The simplest one - (P 1) g(m) + L normally used with a few nodes (bad performance) prediction error < 2% Completion time (s) MPI_Bcast Flat Tree Myrinet Flat Prediction e+060 1e Message size (bytes) Number of nodes LaPIe: Collective Communications adapted to Grid Environments 34 / 60
61 Binomial Tree Broadcast Cost model Broadcast Validating the models log 2 P g(m) + log 2 P L prediction error < 5% Completion time (s) e+060 1e Message size (bytes) MPI_Bcast Binomial Tree Myrinet Binomial Prediction Number of nodes LaPIe: Collective Communications adapted to Grid Environments 35 / 60
62 Segmented Chain Broadcast Cost model Broadcast Validating the models (P 1) (g(s) + L) + (g(s) (k 1)) Performance depends on the segment size Dependent on the performance of all nodes MPI_Bcast Segmented Chain (Pipeline) Myrinet Completion time (s) e+060 1e Message size (bytes) Pipeline 16ko Prediction Number of nodes LaPIe: Collective Communications adapted to Grid Environments 36 / 60
63 Choosing the best strategy Cost model Broadcast Validating the models Comparison MPI_Bcast 25 machines Myrinet Flat Flat prediction Chain Chain prediction Binomial Binomial prediction Completion time (s) e e+06 Message size (bytes) LaPIe: Collective Communications adapted to Grid Environments 37 / 60
64 Cost model Broadcast Validating the models Choosing the best strategy - small messages Comparison MPI_Bcast 25 machines Myrinet Flat Flat prediction Chain Chain prediction Binomial Binomial prediction Completion time (s) Message size (bytes) LaPIe: Collective Communications adapted to Grid Environments 38 / 60
65 Outline Optimising Collective Communications Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation 1 Optimising Collective Communications LaPIe: Collective Communications adapted to Grid Environments 39 / 60
66 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Grid-Aware Collective Communication Scheduling Communications in a Heterogeneous Environment exhaustive search genetic algorithms (Vorakosit) simulated annealing (Vadhiyar) LaPIe: Collective Communications adapted to Grid Environments 40 / 60
67 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Grid-Aware Collective Communication Scheduling Communications in a Heterogeneous Environment exhaustive search genetic algorithms (Vorakosit) simulated annealing (Vadhiyar) operation specific optimisations pipelined broadcasts (Beaumont et al.) balanced trees (Burger et al.). LaPIe: Collective Communications adapted to Grid Environments 40 / 60
68 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Grid-Aware Collective Communication Scheduling Communications in a Heterogeneous Environment exhaustive search genetic algorithms (Vorakosit) simulated annealing (Vadhiyar) operation specific optimisations pipelined broadcasts (Beaumont et al.) balanced trees (Burger et al.). optimisation heuristics FEF and ECEF (Bhat) LaPIe: Collective Communications adapted to Grid Environments 40 / 60
69 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Grid-Aware Collective Communication Why to use an hierarchical scheduling reduces the search space LaPIe: Collective Communications adapted to Grid Environments 41 / 60
70 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Grid-Aware Collective Communication Why to use an hierarchical scheduling reduces the search space each cluster may use different strategies binomial, chain, etc. LaPIe: Collective Communications adapted to Grid Environments 41 / 60
71 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Grid-Aware Collective Communication Why to use an hierarchical scheduling reduces the search space each cluster may use different strategies binomial, chain, etc. this approach may be employed also with other communication patterns LaPIe: Collective Communications adapted to Grid Environments 41 / 60
72 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Broadcast - Optimisation Heuristics Fastest Edge First -FEF (Bhat) objective: select the sender that can reach a new receiver earlier strategy: find the edge with the minimum latency min i A, j B L i,j Drawback this strategy may overload a single sender LaPIe: Collective Communications adapted to Grid Environments 42 / 60
73 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Broadcast - Optimisation Heuristics Earliest Completing Edge First - ECEF (Bhat) objective: select the fastest available sender to reach a new receiver strategy: take into account the Ready Time and the transfer time Weakness (?) min (RT i + g i,j (m) + L i,j ) i A, j B Can the receiver contribute to the broadcast? LaPIe: Collective Communications adapted to Grid Environments 43 / 60
74 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Broadcast - Optimisation Heuristics Earliest Completing Edge First with lookahead - ECEFLA (Bhat) objective: select the fastest available sender to reach a good receiver a node that can contribute with message diffusion strategy: use a lookahead function to evaluate the usefulness of a receiver min (RT i + g i,j (m) + L i,j + F j ) ; i A, j B LaPIe: Collective Communications adapted to Grid Environments 44 / 60
75 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Broadcast - Optimisation Heuristics Earliest Completing Edge First with lookahead - ECEFLA (Bhat) objective: select the fastest available sender to reach a good receiver a node that can contribute with message diffusion strategy: use a lookahead function to evaluate the usefulness of a receiver min (RT i + g i,j (m) + L i,j + F j ) ; F j = min (g j,k(m) + L j,k ) i A, j B P k B LaPIe: Collective Communications adapted to Grid Environments 44 / 60
76 Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Broadcast - Optimisation Heuristics Common characteristics of these heuristics Give priority to fast links Question: maximise the number of potential senders Can a previous knowledge on intra-cluster communications improve the efficiency of these heuristics? T k - communication time inside a cluster LaPIe: Collective Communications adapted to Grid Environments 45 / 60
77 Specific Heuristics Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation ECEFLA-t simple extension of the ECEFLA heuristic objective: select the fastest available sender to reach a good receiver a cluster contacted by this node may finish in the smallest time quickly reduces the number of clusters to contact min (RT i +g i,j (m)+l i,j +F j ) ; F j = min (g j,k(m)+l j,k +T k ) i A, j B P k B LaPIe: Collective Communications adapted to Grid Environments 46 / 60
78 Drawbacks in a Grid System Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation All these strategies always try to contact first the fastest clusters/nodes Communications to distant/slow clusters are delayed This extra delay may augment the makespan Balance communication: Give some priority to slow clusters Still keep trying to reach the largest number of nodes maximise the number of data sources LaPIe: Collective Communications adapted to Grid Environments 47 / 60
79 Specific Heuristics Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation ECEFLA-T - tries to balance the scheduling objective: select a receiver whose cost to contact the slowest cluster is still reduced sender is the fastest one that can reach the slowest cluster strategy: the lookahead function maximises the search min (RT i+g i,j (m)+l i,j +F j ) ; F j = max (g j,k(m)+l j,k +T k ) i A, j B P k B Drawback slow clusters will be contacted only after no fast cluster remains LaPIe: Collective Communications adapted to Grid Environments 48 / 60
80 Specific Heuristics Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Bottom-Up gives priority to slow clusters objective: prevent a supplementary delay for the slow clusters strategy: search for the slowest cluster still not contacted Drawback max (min (g i,j(m) + L i,j + T j )) P j B P i A does not improve the number of data sources LaPIe: Collective Communications adapted to Grid Environments 49 / 60
81 Comparing Strategies Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Simulations Use of simulations to obtain the average performance of each strategy average of runs Random values between: minimum maximum gap i,j 0.10 s 0.60 s IDPOT-icluster2 IDPOT-GdX latency i,j s s IDPOT-icluster2 GdX-Rennes T i 0.02 s 3 s 1 MB Myrinet 1 MB Fast Ethernet LaPIe: Collective Communications adapted to Grid Environments 50 / 60
82 Comparing Strategies Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation 6 5 Flat Tree FEF ECEF ECEF Lookahead ECEFla t ECEFLA T BottomUp 1MB Broadcast in a Grid Environment average of iterations Completion time (s) Number of clusters LaPIe: Collective Communications adapted to Grid Environments 51 / 60
83 A Large Scale Grid Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Flat Tree FEF ECEF ECEF Lookahead ECEFla t ECEFLA T BottomUp 1MB Broadcast in a Grid Environment Average time of iteration Completion time (s) Number of clusters LaPIe: Collective Communications adapted to Grid Environments 52 / 60
84 A Close Look Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation ECEF ECEF Lookahead ECEFla t ECEFla T 1MB Broadcast in a Grid Environment average of iterations Completion time (s) Number of clusters LaPIe: Collective Communications adapted to Grid Environments 53 / 60
85 Hit Rate Optimising Collective Communications Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation A different metric to evaluate the heuristics MB Broadcast in a Grid Environment Hit Rate iterations ECEF ECEF Lookahead ECEFla t ECEFla T Completion time (s) Number of clusters LaPIe: Collective Communications adapted to Grid Environments 54 / 60
86 Experimental validation Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation 88 machines, 6 homogeneous clusters (3 IDPOT,2 GdX, Toulouse) 12 Flat FEF ECEF ECEFLA 10 ECEFLAt ECEFLAT BottomUp Broadcast in a Grid Predicted Times Completion Time (s) e e+06 2e e+06 3e e+06 4e e+06 Message size (Bytes) LaPIe: Collective Communications adapted to Grid Environments 55 / 60
87 Experimental Validation Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Completion time (s) Flat FEF ECEF ECEFLA ECEFLAt ECEFLAT BottomUp Broadcast on a 78 machines grid Measured Times e e+06 2e e+06 3e e+06 4e e+06 Message size (Bytes) LaPIe: Collective Communications adapted to Grid Environments 56 / 60
88 Experimental Validation Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Completion time (s) 12 Flat FEF ECEF ECEFLA 10 ECEFLAt ECEFLAT BottomUp LAM Binomial Broadcast on a 78 machines grid Measured Times e e+06 2e e+06 3e e+06 4e e+06 Message size (Bytes) LaPIe: Collective Communications adapted to Grid Environments 57 / 60
89 Conclusions Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Scheduling communications on a grid environment Hierarchical communication reduces the optimisation complexity Multi-layered communication with hybrid algorithms LaPIe: Collective Communications adapted to Grid Environments 58 / 60
90 Conclusions Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Scheduling communications on a grid environment Hierarchical communication reduces the optimisation complexity Multi-layered communication with hybrid algorithms efficient well known intra-cluster strategies LaPIe: Collective Communications adapted to Grid Environments 58 / 60
91 Conclusions Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Scheduling communications on a grid environment Hierarchical communication reduces the optimisation complexity Multi-layered communication with hybrid algorithms efficient well known intra-cluster strategies dynamically scheduled inter-cluster communications LaPIe: Collective Communications adapted to Grid Environments 58 / 60
92 Conclusions Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Scheduling communications on a grid environment Hierarchical communication reduces the optimisation complexity Multi-layered communication with hybrid algorithms efficient well known intra-cluster strategies dynamically scheduled inter-cluster communications Importance of Topology Discovery Helps to better describe the real network Prevents mistakes induced by manual configuration Simplify further optimisation tasks LaPIe: Collective Communications adapted to Grid Environments 58 / 60
93 Future Works Scheduling Strategies Optimisation Heuristics Simulation Experimental Validation Extend our experiments More experiments on a grid environment Compare with other heuristics and optimisation techniques Evaluate the impact on the performance of real applications LaPIe: Collective Communications adapted to Grid Environments 59 / 60
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,
A Review of Customized Dynamic Load Balancing for a Network of Workstations
A Review of Customized Dynamic Load Balancing for a Network of Workstations Taken from work done by: Mohammed Javeed Zaki, Wei Li, Srinivasan Parthasarathy Computer Science Department, University of Rochester
Clustering UE 141 Spring 2013
Clustering UE 141 Spring 013 Jing Gao SUNY Buffalo 1 Definition of Clustering Finding groups of obects such that the obects in a group will be similar (or related) to one another and different from (or
Traffic Engineering for Multiple Spanning Tree Protocol in Large Data Centers
Traffic Engineering for Multiple Spanning Tree Protocol in Large Data Centers Ho Trong Viet, Yves Deville, Olivier Bonaventure, Pierre François ICTEAM, Université catholique de Louvain (UCL), Belgium.
Load Balancing in Periodic Wireless Sensor Networks for Lifetime Maximisation
Load Balancing in Periodic Wireless Sensor Networks for Lifetime Maximisation Anthony Kleerekoper 2nd year PhD Multi-Service Networks 2011 The Energy Hole Problem Uniform distribution of motes Regular,
Interconnection Networks. Interconnection Networks. Interconnection networks are used everywhere!
Interconnection Networks Interconnection Networks Interconnection networks are used everywhere! Supercomputers connecting the processors Routers connecting the ports can consider a router as a parallel
- An Essential Building Block for Stable and Reliable Compute Clusters
Ferdinand Geier ParTec Cluster Competence Center GmbH, V. 1.4, March 2005 Cluster Middleware - An Essential Building Block for Stable and Reliable Compute Clusters Contents: Compute Clusters a Real Alternative
Dynamic Load Balancing in Charm++ Abhinav S Bhatele Parallel Programming Lab, UIUC
Dynamic Load Balancing in Charm++ Abhinav S Bhatele Parallel Programming Lab, UIUC Outline Dynamic Load Balancing framework in Charm++ Measurement Based Load Balancing Examples: Hybrid Load Balancers Topology-aware
A Peer-to-peer Extension of Network-Enabled Server Systems
A Peer-to-peer Extension of Network-Enabled Server Systems Eddy Caron 1, Frédéric Desprez 1, Cédric Tedeschi 1 Franck Petit 2 1 - GRAAL Project / LIP laboratory 2 - LaRIA laboratory E-Science 2005 - December
An MPI Tool for Automatically Discovering the Switch Level Topologies of Ethernet Clusters
An MPI Tool for Automatically Discovering the Switch Level Topologies of Ethernet Clusters Joshua Lawrence Xin Yuan Department of Computer Science, Florida State University, Tallahassee, FL 32306 {lawrence,
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.
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
Performance Monitoring of Parallel Scientific Applications
Performance Monitoring of Parallel Scientific Applications Abstract. David Skinner National Energy Research Scientific Computing Center Lawrence Berkeley National Laboratory This paper introduces an infrastructure
A Simultaneous Solution for General Linear Equations on a Ring or Hierarchical Cluster
Acta Technica Jaurinensis Vol. 3. No. 1. 010 A Simultaneous Solution for General Linear Equations on a Ring or Hierarchical Cluster G. Molnárka, N. Varjasi Széchenyi István University Győr, Hungary, H-906
Ethernet. Ethernet Frame Structure. Ethernet Frame Structure (more) Ethernet: uses CSMA/CD
Ethernet dominant LAN technology: cheap -- $20 for 100Mbs! first widely used LAN technology Simpler, cheaper than token rings and ATM Kept up with speed race: 10, 100, 1000 Mbps Metcalfe s Etheret sketch
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
Distributed RAID Architectures for Cluster I/O Computing. Kai Hwang
Distributed RAID Architectures for Cluster I/O Computing Kai Hwang Internet and Cluster Computing Lab. University of Southern California 1 Presentation Outline : Scalable Cluster I/O The RAID-x Architecture
Clustering. 15-381 Artificial Intelligence Henry Lin. Organizing data into clusters such that there is
Clustering 15-381 Artificial Intelligence Henry Lin Modified from excellent slides of Eamonn Keogh, Ziv Bar-Joseph, and Andrew Moore What is Clustering? Organizing data into clusters such that there is
A NOVEL RESOURCE EFFICIENT DMMS APPROACH
A NOVEL RESOURCE EFFICIENT DMMS APPROACH FOR NETWORK MONITORING AND CONTROLLING FUNCTIONS Golam R. Khan 1, Sharmistha Khan 2, Dhadesugoor R. Vaman 3, and Suxia Cui 4 Department of Electrical and Computer
Distributed communication-aware load balancing with TreeMatch in Charm++
Distributed communication-aware load balancing with TreeMatch in Charm++ The 9th Scheduling for Large Scale Systems Workshop, Lyon, France Emmanuel Jeannot Guillaume Mercier Francois Tessier In collaboration
EonStor DS remote replication feature guide
EonStor DS remote replication feature guide White paper Version: 1.0 Updated: Abstract: Remote replication on select EonStor DS storage systems offers strong defense against major disruption to IT continuity,
Bandwidth Efficient All-to-All Broadcast on Switched Clusters
Bandwidth Efficient All-to-All Broadcast on Switched Clusters Ahmad Faraj Pitch Patarasuk Xin Yuan Blue Gene Software Development Department of Computer Science IBM Corporation Florida State University
Network Architecture and Topology
1. Introduction 2. Fundamentals and design principles 3. Network architecture and topology 4. Network control and signalling 5. Network components 5.1 links 5.2 switches and routers 6. End systems 7. End-to-end
Enabling Large-Scale Testing of IaaS Cloud Platforms on the Grid 5000 Testbed
Enabling Large-Scale Testing of IaaS Cloud Platforms on the Grid 5000 Testbed Sébastien Badia, Alexandra Carpen-Amarie, Adrien Lèbre, Lucas Nussbaum Grid 5000 S. Badia, A. Carpen-Amarie, A. Lèbre, L. Nussbaum
Load Balancing in Distributed Web Server Systems With Partial Document Replication
Load Balancing in Distributed Web Server Systems With Partial Document Replication Ling Zhuo, Cho-Li Wang and Francis C. M. Lau Department of Computer Science and Information Systems The University of
Distributed Systems LEEC (2005/06 2º Sem.)
Distributed Systems LEEC (2005/06 2º Sem.) Introduction João Paulo Carvalho Universidade Técnica de Lisboa / Instituto Superior Técnico Outline Definition of a Distributed System Goals Connecting Users
Parallel Computing. Benson Muite. benson.muite@ut.ee http://math.ut.ee/ benson. https://courses.cs.ut.ee/2014/paralleel/fall/main/homepage
Parallel Computing Benson Muite benson.muite@ut.ee http://math.ut.ee/ benson https://courses.cs.ut.ee/2014/paralleel/fall/main/homepage 3 November 2014 Hadoop, Review Hadoop Hadoop History Hadoop Framework
High Performance Computing. Course Notes 2007-2008. HPC Fundamentals
High Performance Computing Course Notes 2007-2008 2008 HPC Fundamentals Introduction What is High Performance Computing (HPC)? Difficult to define - it s a moving target. Later 1980s, a supercomputer performs
PeerMon: A Peer-to-Peer Network Monitoring System
PeerMon: A Peer-to-Peer Network Monitoring System Tia Newhall, Janis Libeks, Ross Greenwood, Jeff Knerr Computer Science Department Swarthmore College Swarthmore, PA USA newhall@cs.swarthmore.edu Target:
IP Multicasting. Applications with multiple receivers
IP Multicasting Relates to Lab 10. It covers IP multicasting, including multicast addressing, IGMP, and multicast routing. 1 Applications with multiple receivers Many applications transmit the same data
How To Monitor Infiniband Network Data From A Network On A Leaf Switch (Wired) On A Microsoft Powerbook (Wired Or Microsoft) On An Ipa (Wired/Wired) Or Ipa V2 (Wired V2)
INFINIBAND NETWORK ANALYSIS AND MONITORING USING OPENSM N. Dandapanthula 1, H. Subramoni 1, J. Vienne 1, K. Kandalla 1, S. Sur 1, D. K. Panda 1, and R. Brightwell 2 Presented By Xavier Besseron 1 Date:
Definition. A Historical Example
Overlay Networks This lecture contains slides created by Ion Stoica (UC Berkeley). Slides used with permission from author. All rights remain with author. Definition Network defines addressing, routing,
Scalability and Classifications
Scalability and Classifications 1 Types of Parallel Computers MIMD and SIMD classifications shared and distributed memory multicomputers distributed shared memory computers 2 Network Topologies static
COMP 422, Lecture 3: Physical Organization & Communication Costs in Parallel Machines (Sections 2.4 & 2.5 of textbook)
COMP 422, Lecture 3: Physical Organization & Communication Costs in Parallel Machines (Sections 2.4 & 2.5 of textbook) Vivek Sarkar Department of Computer Science Rice University vsarkar@rice.edu COMP
Information Processing, Big Data, and the Cloud
Information Processing, Big Data, and the Cloud James Horey Computational Sciences & Engineering Oak Ridge National Laboratory Fall Creek Falls 2010 Information Processing Systems Model Parameters Data-intensive
Kerrighed / XtreemOS cluster flavour
Kerrighed / XtreemOS cluster flavour Jean Parpaillon Reisensburg Castle Günzburg, Germany July 5-9, 2010 July 6th, 2010 Kerrighed - XtreemOS cluster flavour 1 Summary Kerlabs Context Kerrighed Project
Load Balancing Support for Grid-enabled Applications
John von Neumann Institute for Computing Load Balancing Support for Grid-enabled Applications S. Rips published in Parallel Computing: Current & Future Issues of High-End Computing, Proceedings of the
PROFINET the Industrial Ethernet standard. Siemens AG 2013. Alle Rechte vorbehalten.
the Industrial Ethernet standard is 100% Ethernet is Ethernet Ethernet is the established standard in the IT world for fast exchange of data (IEEE 802.3) is always full duplex simultaneous communication
Performance Characteristics of a Cost-Effective Medium-Sized Beowulf Cluster Supercomputer
Res. Lett. Inf. Math. Sci., 2003, Vol.5, pp 1-10 Available online at http://iims.massey.ac.nz/research/letters/ 1 Performance Characteristics of a Cost-Effective Medium-Sized Beowulf Cluster Supercomputer
Dynamic Network Resources Allocation in Grids through a Grid Network Resource Broker
INGRID 2007 Instrumenting the GRID Second International Workshop on Distributed Cooperative Laboratories Session 2: Networking for the GRID Dynamic Network Resources Allocation in Grids through a Grid
Rodrigo Fernandes de Mello, Evgueni Dodonov, José Augusto Andrade Filho
Middleware for High Performance Computing Rodrigo Fernandes de Mello, Evgueni Dodonov, José Augusto Andrade Filho University of São Paulo São Carlos, Brazil {mello, eugeni, augustoa}@icmc.usp.br Outline
Introduction Page 2. Understanding Bandwidth Units Page 3. Internet Bandwidth V/s Download Speed Page 4. Optimum Utilization of Bandwidth Page 8
INDEX Introduction Page 2 Understanding Bandwidth Units Page 3 Internet Bandwidth V/s Download Speed Page 4 Factors Affecting Download Speed Page 5-7 Optimum Utilization of Bandwidth Page 8 Conclusion
Cluster Analysis. Isabel M. Rodrigues. Lisboa, 2014. Instituto Superior Técnico
Instituto Superior Técnico Lisboa, 2014 Introduction: Cluster analysis What is? Finding groups of objects such that the objects in a group will be similar (or related) to one another and different from
A SIMULATOR FOR LOAD BALANCING ANALYSIS IN DISTRIBUTED SYSTEMS
Mihai Horia Zaharia, Florin Leon, Dan Galea (3) A Simulator for Load Balancing Analysis in Distributed Systems in A. Valachi, D. Galea, A. M. Florea, M. Craus (eds.) - Tehnologii informationale, Editura
ENABLING THE PRIVATE CLOUD - THE NEW DATA CENTER NETWORK. David Yen EVP and GM, Fabric and Switching Technologies Juniper Networks
ENABLING THE PRIVATE CLOUD - THE NEW DATA CENTER NETWORK David Yen EVP and GM, Fabric and Switching Technologies Juniper Networks Services delivered over the Network Dynamically shared resource pools Application
Expanding the CASEsim Framework to Facilitate Load Balancing of Social Network Simulations
Expanding the CASEsim Framework to Facilitate Load Balancing of Social Network Simulations Amara Keller, Martin Kelly, Aaron Todd 4 June 2010 Abstract This research has two components, both involving the
Unsupervised learning: Clustering
Unsupervised learning: Clustering Salissou Moutari Centre for Statistical Science and Operational Research CenSSOR 17 th September 2013 Unsupervised learning: Clustering 1/52 Outline 1 Introduction What
HCOC: A Cost Optimization Algorithm for Workflow Scheduling in Hybrid Clouds
Noname manuscript No. (will be inserted by the editor) : A Cost Optimization Algorithm for Workflow Scheduling in Hybrid Clouds Luiz Fernando Bittencourt Edmundo Roberto Mauro Madeira Received: date /
PART II. OPS-based metro area networks
PART II OPS-based metro area networks Chapter 3 Introduction to the OPS-based metro area networks Some traffic estimates for the UK network over the next few years [39] indicate that when access is primarily
Dynamic Load Balancing Strategy for Grid Computing
Dynamic Load Balancing Strategy for Grid Computing Belabbas Yagoubi and Yahya Slimani Abstract Workload and resource management are two essential functions provided at the service level of the grid software
In-Situ Bitmaps Generation and Efficient Data Analysis based on Bitmaps. Yu Su, Yi Wang, Gagan Agrawal The Ohio State University
In-Situ Bitmaps Generation and Efficient Data Analysis based on Bitmaps Yu Su, Yi Wang, Gagan Agrawal The Ohio State University Motivation HPC Trends Huge performance gap CPU: extremely fast for generating
CHAPTER 5 WLDMA: A NEW LOAD BALANCING STRATEGY FOR WAN ENVIRONMENT
81 CHAPTER 5 WLDMA: A NEW LOAD BALANCING STRATEGY FOR WAN ENVIRONMENT 5.1 INTRODUCTION Distributed Web servers on the Internet require high scalability and availability to provide efficient services to
Computer Networking: A Survey
Computer Networking: A Survey M. Benaiah Deva Kumar and B. Deepa, 1 Scholar, 2 Assistant Professor, IT Department, Sri Krishna College of Arts and Science College, Coimbatore, India. Abstract- Computer
Efficient Crawling of Community Structures in Online Social Networks
Efficient Crawling of Community Structures in Online Social Networks Network Architectures and Services PVM 2011-071 Efficient Crawling of Community Structures in Online Social Networks For the degree
MPI Implementation Analysis - A Practical Approach to Network Marketing
Optimizing MPI Collective Communication by Orthogonal Structures Matthias Kühnemann Fakultät für Informatik Technische Universität Chemnitz 917 Chemnitz, Germany kumat@informatik.tu chemnitz.de Gudula
On real-time delay monitoring in software-defined networks
On real-time delay monitoring in software-defined networks Victor S. Altukhov Lomonosov Moscow State University Moscow, Russia victoralt@lvk.cs.msu.su Eugene V. Chemeritskiy Applied Research Center for
Latency on a Switched Ethernet Network
Application Note 8 Latency on a Switched Ethernet Network Introduction: This document serves to explain the sources of latency on a switched Ethernet network and describe how to calculate cumulative latency
Praktikum Wissenschaftliches Rechnen (Performance-optimized optimized Programming)
Praktikum Wissenschaftliches Rechnen (Performance-optimized optimized Programming) Dynamic Load Balancing Dr. Ralf-Peter Mundani Center for Simulation Technology in Engineering Technische Universität München
Interconnect Efficiency of Tyan PSC T-630 with Microsoft Compute Cluster Server 2003
Interconnect Efficiency of Tyan PSC T-630 with Microsoft Compute Cluster Server 2003 Josef Pelikán Charles University in Prague, KSVI Department, Josef.Pelikan@mff.cuni.cz Abstract 1 Interconnect quality
On the Placement of Management and Control Functionality in Software Defined Networks
On the Placement of Management and Control Functionality in Software Defined Networks D.Tuncer et al. Department of Electronic & Electrical Engineering University College London, UK ManSDN/NfV 13 November
Computer Networks. Definition of LAN. Connection of Network. Key Points of LAN. Lecture 06 Connecting Networks
Computer Networks Lecture 06 Connecting Networks Kuang-hua Chen Department of Library and Information Science National Taiwan University Local Area Networks (LAN) 5 kilometer IEEE 802.3 Ethernet IEEE 802.4
Efficient and Robust Allocation Algorithms in Clouds under Memory Constraints
Efficient and Robust Allocation Algorithms in Clouds under Memory Constraints Olivier Beaumont,, Paul Renaud-Goud Inria & University of Bordeaux Bordeaux, France 9th Scheduling for Large Scale Systems
Axon: A Flexible Substrate for Source- routed Ethernet. Jeffrey Shafer Brent Stephens Michael Foss Sco6 Rixner Alan L. Cox
Axon: A Flexible Substrate for Source- routed Ethernet Jeffrey Shafer Brent Stephens Michael Foss Sco6 Rixner Alan L. Cox 2 Ethernet Tradeoffs Strengths Weaknesses Cheap Simple High data rate Ubiquitous
Load Balancing Between Heterogenous Computing Clusters
Load Balancing Between Heterogenous Computing Clusters Siu-Cheung Chau Dept. of Physics and Computing, Wilfrid Laurier University, Waterloo, Ontario, Canada, N2L 3C5 e-mail: schau@wlu.ca Ada Wai-Chee Fu
Windows Server Performance Monitoring
Spot server problems before they are noticed The system s really slow today! How often have you heard that? Finding the solution isn t so easy. The obvious questions to ask are why is it running slowly
Aggregate Router: An Efficient Inter-Cluster MPI Communication Facility
IPSJ Online Transactions Vol. 2 215 224 (Oct. 2009) Regular Paper Aggregate Router: An Efficient Inter-Cluster MPI Communication Facility Hiroya Matsuba 1 and Yutaka Ishikawa 2 At a cluster of clusters
CSE 4351/5351 Notes 7: Task Scheduling & Load Balancing
CSE / Notes : Task Scheduling & Load Balancing Task Scheduling A task is a (sequential) activity that uses a set of inputs to produce a set of outputs. A task (precedence) graph is an acyclic, directed
Performance Analysis and Optimization Tool
Performance Analysis and Optimization Tool Andres S. CHARIF-RUBIAL andres.charif@uvsq.fr Performance Analysis Team, University of Versailles http://www.maqao.org Introduction Performance Analysis Develop
Cluster, Grid, Cloud Concepts
Cluster, Grid, Cloud Concepts Kalaiselvan.K Contents Section 1: Cluster Section 2: Grid Section 3: Cloud Cluster An Overview Need for a Cluster Cluster categorizations A computer cluster is a group of
A Flexible Cluster Infrastructure for Systems Research and Software Development
Award Number: CNS-551555 Title: CRI: Acquisition of an InfiniBand Cluster with SMP Nodes Institution: Florida State University PIs: Xin Yuan, Robert van Engelen, Kartik Gopalan A Flexible Cluster Infrastructure
CORPORATE NETWORKING
CORPORATE NETWORKING C. Pham Université de Pau et des Pays de l Adour Département Informatique http://www.univ-pau.fr/~cpham Congduc.Pham@univ-pau.fr Typical example of Ethernet local networks Mostly based
An Efficient Hybrid Data Gathering Scheme in Wireless Sensor Networks
An Efficient Hybrid Data Gathering Scheme in Wireless Sensor Networks Ayon Chakraborty 1, Swarup Kumar Mitra 2, and M.K. Naskar 3 1 Department of CSE, Jadavpur University, Kolkata, India 2 Department of
MPI / ClusterTools Update and Plans
HPC Technical Training Seminar July 7, 2008 October 26, 2007 2 nd HLRS Parallel Tools Workshop Sun HPC ClusterTools 7+: A Binary Distribution of Open MPI MPI / ClusterTools Update and Plans Len Wisniewski
The Feasibility of Supporting Large-Scale Live Streaming Applications with Dynamic Application End-Points
The Feasibility of Supporting Large-Scale Live Streaming Applications with Dynamic Application End-Points Kay Sripanidkulchai, Aditya Ganjam, Bruce Maggs, and Hui Zhang Instructor: Fabian Bustamante Presented
Advancing Applications Performance With InfiniBand
Advancing Applications Performance With InfiniBand Pak Lui, Application Performance Manager September 12, 2013 Mellanox Overview Ticker: MLNX Leading provider of high-throughput, low-latency server and
Overlapping Data Transfer With Application Execution on Clusters
Overlapping Data Transfer With Application Execution on Clusters Karen L. Reid and Michael Stumm reid@cs.toronto.edu stumm@eecg.toronto.edu Department of Computer Science Department of Electrical and Computer
HPAM: Hybrid Protocol for Application Level Multicast. Yeo Chai Kiat
HPAM: Hybrid Protocol for Application Level Multicast Yeo Chai Kiat Scope 1. Introduction 2. Hybrid Protocol for Application Level Multicast (HPAM) 3. Features of HPAM 4. Conclusion 1. Introduction Video
Introduction to LAN/WAN. Network Layer
Introduction to LAN/WAN Network Layer Topics Introduction (5-5.1) Routing (5.2) (The core) Internetworking (5.5) Congestion Control (5.3) Network Layer Design Isues Store-and-Forward Packet Switching Services
1.1 Difficulty in Fault Localization in Large-Scale Computing Systems
Chapter 1 Introduction System failures have been one of the biggest obstacles in operating today s largescale computing systems. Fault localization, i.e., identifying direct or indirect causes of failures,
Scientific application deployment on Cloud: A Topology-Aware Method
CONCURRENCY AND COMPUTATION: PRACTICE AND EXPERIENCE Concurrency Computat.: Pract. Exper. 212; :1 2 Published online in Wiley InterScience (www.interscience.wiley.com). Scientific application deployment
Contents. Foreword. Acknowledgments
Foreword Preface Acknowledgments xv xvii xviii CHAPTER 1 Introduction 1 1.1 What Is Mission Critical? 1 1.2 Purpose of the Book 2 1.3 Network Continuity Versus Disaster Recovery 2 1.4 The Case for Mission-Critical
Multi-Channel Clustered Web Application Servers
THE AMERICAN UNIVERSITY IN CAIRO SCHOOL OF SCIENCES AND ENGINEERING Multi-Channel Clustered Web Application Servers A Masters Thesis Department of Computer Science and Engineering Status Report Seminar
Big Data Processing with Google s MapReduce. Alexandru Costan
1 Big Data Processing with Google s MapReduce Alexandru Costan Outline Motivation MapReduce programming model Examples MapReduce system architecture Limitations Extensions 2 Motivation Big Data @Google:
Study on Cloud Computing Resource Scheduling Strategy Based on the Ant Colony Optimization Algorithm
www.ijcsi.org 54 Study on Cloud Computing Resource Scheduling Strategy Based on the Ant Colony Optimization Algorithm Linan Zhu 1, Qingshui Li 2, and Lingna He 3 1 College of Mechanical Engineering, Zhejiang
Scheduling Allowance Adaptability in Load Balancing technique for Distributed Systems
Scheduling Allowance Adaptability in Load Balancing technique for Distributed Systems G.Rajina #1, P.Nagaraju #2 #1 M.Tech, Computer Science Engineering, TallaPadmavathi Engineering College, Warangal,
Optimization in Content Distribution Networks
EngOpt 2008 - International Conference on Engineering Optimization Rio de Janeiro, Brazil, 01-05 June 2008. Optimization in Content Distribution Networks Tiago Araújo Neves, Luiz Satoru Ochi, Lúcia M.
Storage Systems Autumn 2009. Chapter 6: Distributed Hash Tables and their Applications André Brinkmann
Storage Systems Autumn 2009 Chapter 6: Distributed Hash Tables and their Applications André Brinkmann Scaling RAID architectures Using traditional RAID architecture does not scale Adding news disk implies
Performance of the NAS Parallel Benchmarks on Grid Enabled Clusters
Performance of the NAS Parallel Benchmarks on Grid Enabled Clusters Philip J. Sokolowski Dept. of Electrical and Computer Engineering Wayne State University 55 Anthony Wayne Dr., Detroit, MI 4822 phil@wayne.edu
Distributed applications monitoring at system and network level
Distributed applications monitoring at system and network level Monarc Collaboration 1 Abstract Most of the distributed applications are presently based on architectural models that don t involve real-time
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,,
QUALITY OF SERVICE METRICS FOR DATA TRANSMISSION IN MESH TOPOLOGIES
QUALITY OF SERVICE METRICS FOR DATA TRANSMISSION IN MESH TOPOLOGIES SWATHI NANDURI * ZAHOOR-UL-HUQ * Master of Technology, Associate Professor, G. Pulla Reddy Engineering College, G. Pulla Reddy Engineering
Load Balancing Mechanisms in Data Center Networks
Load Balancing Mechanisms in Data Center Networks Santosh Mahapatra Xin Yuan Department of Computer Science, Florida State University, Tallahassee, FL 33 {mahapatr,xyuan}@cs.fsu.edu Abstract We consider
Overview of Network Hardware and Software. CS158a Chris Pollett Jan 29, 2007.
Overview of Network Hardware and Software CS158a Chris Pollett Jan 29, 2007. Outline Scales of Networks Protocol Hierarchies Scales of Networks Last day, we talked about broadcast versus point-to-point
Datagram-based network layer: forwarding; routing. Additional function of VCbased network layer: call setup.
CEN 007C Computer Networks Fundamentals Instructor: Prof. A. Helmy Homework : Network Layer Assigned: Nov. 28 th, 2011. Due Date: Dec 8 th, 2011 (to the TA) 1. ( points) What are the 2 most important network-layer
Lecture 4: Introduction to Computer Network Design
Lecture 4: Introduction to Computer Design Prof. Shervin Shirmohammadi SITE, University of Ottawa Prof. Shervin Shirmohammadi CEG 4185 4-1 Computer s Prof. Shervin Shirmohammadi CEG 4185 4-2 1 Background
CROSS LAYER BASED MULTIPATH ROUTING FOR LOAD BALANCING
CHAPTER 6 CROSS LAYER BASED MULTIPATH ROUTING FOR LOAD BALANCING 6.1 INTRODUCTION The technical challenges in WMNs are load balancing, optimal routing, fairness, network auto-configuration and mobility
Proceedings of the Federated Conference on Computer Science and Information Systems pp. 737 741
Proceedings of the Federated Conference on Computer Science and Information Systems pp. 737 741 ISBN 978-83-60810-22-4 DCFMS: A Chunk-Based Distributed File System for Supporting Multimedia Communication
http://www.paper.edu.cn
5 10 15 20 25 30 35 A platform for massive railway information data storage # SHAN Xu 1, WANG Genying 1, LIU Lin 2** (1. Key Laboratory of Communication and Information Systems, Beijing Municipal Commission
Fairness issues in new large scale parallel platforms.
Fairness issues in new large scale parallel platforms. Denis TRYSTRAM LIG Université de Grenoble Alpes Inria Institut Universitaire de France july 5, 25 New computing systems New challenges from e-science
Cluster Computing at HRI
Cluster Computing at HRI J.S.Bagla Harish-Chandra Research Institute, Chhatnag Road, Jhunsi, Allahabad 211019. E-mail: jasjeet@mri.ernet.in 1 Introduction and some local history High performance computing