Agent-Based Grid Load-Balancing

Daniel P. SpoonerUniversity of Warwick, UK

Junwei CaoNEC Europe Ltd., Germany

Outline Grid Computing & Middleware Agent-Based Methodology Distributed Service Discovery Grid Load-Balancing Experimental Results Conclusions & Future Work

Grid Computing Computational Grids - blueprint for a

new computing infrastructure

Grid/Web Services - distributed system and application integration

Data/Service/Community Grids - large-scale resource sharing in VOs

Grid Middleware Resource Management & Scheduling Information Services (Monitoring & Discovery) Data Management and Access Application Programming Environments Security, Accounting, QoS ……

Condor, LSF, Ninf, Nimrod, ……Globus, Legion, DPSS, ……Java/Jini, CORBA, Web Services, ……

Grid Resource Management

Key challenges: Cross-domain Large-scale Dynamic QoS support

GRM

Agent-Based MethodologyAn agent is: A local grid manager An user agent A broker A service provider A service requestor A matchmaker

AAA A

Local ManagementProcessor 1Processor 2Processor 3Processor 4Processor 5Processor 6Processor 7Processor 8

2n-1

Performance Prediction Task models Hardware models

FIFO AlgorithmGenetic Algorithm Heuristic &

Evolutionary Near-optimal on

makespan, deadlines and idletime.

Service Discovery Pure data-pull No advertisement Full discovery Efficient when service

change more quickly

Pure data-push Full advertisement No discovery Efficient when

requests arrive more frequently

R/A

A U/A1 TaskInfo

2 PullSerInfo

3 PullSerInfo4 Return

R/A

R/A

A U/A3 TaskInfo

1 PushSerInfo

2 PushSerInfo4 Return

R/A

Centralised, not applicable for grid computing!

Optimisation Strategies Configurable data-pull or data-push Agent hierarchy Multi-step advertisement & multi-step discovery Efficient when frequencies of request arrivals

and service changes are almost the same

Distributed!balancing betweenadvertisement anddiscovery!

A

A A

A A

User

Agent Implementation

PerformancePrediction

FIFOScheduling

DatabaseManagement

MatchMaker

Advertisement Discovery

Task Model

Results

Sched. Time User Deadline

To Another Agent

Task Execution

TaskManagement

FIFO/GA Scheduling

Service Info.Provider

Current M

akespanService Info

Hw Model

Task ExecutionResourceMonitoring

Task Execution

Load Balancing Metrics Total makespan Average advance time of task execution

completions (required deadline - actual task completion time)

Average processor utilisation rate (busy time / total makespan)

Load balancing level (1 - mean square deviation of processor utilisation rates / average processor utilisation rate)

Total number of network packages for both advertisement and discovery

Experiment DesignS1

(SGIOrigin2000, 16)

S2(SGIOrigin2000, 16)

S4(SunUltra10,

16)S3(SunUltra10,

16)

S5(SunUltra5,

16)

S6(SunUltra5,

16)

S12(SunSPARCsta

tion2, 16)

S11(SunSPARCsta

tion2, 16)

S8(SunUltra1,

16)

S7(SunUltra5,

16)

S10(SunUltra1,

16)

S9(SunUltra1,

16)

Tasks:sweep3dfftimprocclosurejacobimemsortcpi

Experiment 1 FIFO

FIFO FIFO

FIFO FIFO

Experiment 2 GA

GA GA

GA GA

Experiment 3 GA

GA GA

GA GA

Task Execution

-1200

-1000

-800

-600

-400

-200

0

200

1 2 3

Experiment Number

ε (s

)

S1S2S3S4S5S6S7S8S9S10S11S12Total

Both GA and agents contribute towards the improvement in task executions.

Resource Utilisation

0

10

20

30

40

50

60

70

80

90

1 2 3

Experiment Number

υ (%

)

S1S2S3S4S5S6S7S8S9S10S11S12Total

Less powerful S11 & S12 benefit mainly from the GA.

More powerful S1 & S2 benefit mainly from agents.

Load Balancing

0

10

20

30

40

50

60

70

80

90

100

1 2 3

Experiment Number

β (%

)

S1S2S3S4S5S6S7S8S9S10S11S12Total

The GA contributes more to local grid load balancing.

Agents contribute more to global grid load balancing.

Total Makespan

The centralised pure data-pull can always achieve the best results

Distributed agents with the hierarchical model can also achieve reasonably good results0

20

4060

80100

120140

160

4 6 8 10 12 14 16 18 20

Agent Number

t (m

ins)

CentralisedDistributed

Network Package

The network overhead for the pure data-pull strategy to achieve better results is very high.

Distributed agent-based service advertisement and discovery can scale well.

0

5000

10000

15000

20000

25000

4 6 8 10 12 14 16 18 20

Agent Number

pack

ets

CentralisedDistributed

Conclusions An multi-agent paradigm provides a

clear high-level abstraction of grid resource management system.

Distributed service advertisement and discovery strategies can be used to improve agent performance.

Agent-based framework is scalable, flexible, and extensible for further enhancements.