Proceedings of the seventeenth ACM symposium on Operating systems principles
ACM Transactions on Computer Systems (TOCS)
A Demand Adaptive and Locality Aware (DALA) streaming media server cluster architecture
NOSSDAV '02 Proceedings of the 12th international workshop on Network and operating systems support for digital audio and video
Improving the scalability of the CORBA event service with a multi-agent load balancing algorithm
Software—Practice & Experience
Grid resource management
SEcS: scalable edge-computing services
Proceedings of the 2005 ACM symposium on Applied computing
Using random subsets to build scalable network services
USITS'03 Proceedings of the 4th conference on USENIX Symposium on Internet Technologies and Systems - Volume 4
Load sharing in Call Server clusters
Computer Communications
Compact samples for data dissemination
Journal of Computer and System Sciences
Compact samples for data dissemination
ICDT'07 Proceedings of the 11th international conference on Database Theory
Hi-index | 0.00 |
In this paper we examine the problem of balancing load in a large-scale distributed system when information about server loads may be stale. It is well known that sending each request to the machine with the apparent lowest load can behave badly in such systems, yet this technique is common in practice. Other systems use round-robin or random selection algorithms that entirely ignore load information or that only use a small subset of the load information. Rather than risk extremely bad performance on one hand or ignore the chance to use load information to improve performance on the other, we develop strategies that interpret load information based on its age. Through simulation, we examine several simple algorithms that use such load interpretation strategies under a range of workloads. Our experiments suggest that by properly interpreting load information, systems can (1) match the performance of the most aggressive algorithms when load information is fresh relative to the job arrival rate, (2) outperform the best of the other algorithms we examine by as much as 60% when information is moderately old, (3) significantly outperform random load distribution when information is older still, and (4) avoid pathological behavior even when information is extremely old.