TY - GEN
T1 - An agent based dynamic load balancing system
AU - Rajagopalan, A.
AU - Hariri, S.
N1 - Publisher Copyright:
© 2000 IEEE.
PY - 2000
Y1 - 2000
N2 - High-end workstations being immensely underutilized and a selected few being overloaded reflects on the poor performance of a cluster. Load balancing, assigning each processor workload proportional to its performance capability, could significantly enhance the resource utilization to cost ratio of a cluster, and hence reduce the overall execution time of the clusters' processes. In this paper we present an agent-based dynamic load balancing framework for heterogeneous clusters of computing systems. The Dynamic Agent System for a Heterogeneous cluster (DASH) is a middle tier as architecture above the system level which dynamically provides for n tasks non-preemptive task scheduling, application handling, and fault tolerance. Our approach dynamically configures/constructs the load balancing scheme depending on the current state of the heterogeneous cluster. DASH services are implemented using agents running on each node that collaborate dynamically to establish a global awareness of the system resources and states. Based on this dynamic global awareness, we use a combination of load metrics and statistical predication metrics to schedule processes and thus balance the loads across all the clusters of computers. Our preliminary experimental results for various test cases with different combinations of load metrics are analyzed to show the performance gains that can be achieved by DASH.
AB - High-end workstations being immensely underutilized and a selected few being overloaded reflects on the poor performance of a cluster. Load balancing, assigning each processor workload proportional to its performance capability, could significantly enhance the resource utilization to cost ratio of a cluster, and hence reduce the overall execution time of the clusters' processes. In this paper we present an agent-based dynamic load balancing framework for heterogeneous clusters of computing systems. The Dynamic Agent System for a Heterogeneous cluster (DASH) is a middle tier as architecture above the system level which dynamically provides for n tasks non-preemptive task scheduling, application handling, and fault tolerance. Our approach dynamically configures/constructs the load balancing scheme depending on the current state of the heterogeneous cluster. DASH services are implemented using agents running on each node that collaborate dynamically to establish a global awareness of the system resources and states. Based on this dynamic global awareness, we use a combination of load metrics and statistical predication metrics to schedule processes and thus balance the loads across all the clusters of computers. Our preliminary experimental results for various test cases with different combinations of load metrics are analyzed to show the performance gains that can be achieved by DASH.
KW - Application-level
KW - Collaborative agents
KW - Distributed node system
KW - Dynamic Load-balancing
UR - http://www.scopus.com/inward/record.url?scp=84962756072&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84962756072&partnerID=8YFLogxK
U2 - 10.1109/IWADS.2000.880903
DO - 10.1109/IWADS.2000.880903
M3 - Conference contribution
AN - SCOPUS:84962756072
T3 - Proceedings - 2000 International Workshop on Autonomous Decentralized System
SP - 164
EP - 171
BT - Proceedings - 2000 International Workshop on Autonomous Decentralized System
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - International Workshop on Autonomous Decentralized System
Y2 - 21 September 2000 through 23 September 2000
ER -