A comparison of receiver-initiated and sender-initiated adaptive load sharing
Performance Evaluation
The Power of Two Choices in Randomized Load Balancing
IEEE Transactions on Parallel and Distributed Systems
A mean field model of work stealing in large-scale systems
Proceedings of the ACM SIGMETRICS international conference on Measurement and modeling of computer systems
| Hi-index | 0.00 |
Consider a randomized load balancing problem consisting of a large number n of server sites each equipped with K servers. Under the greedy policy, clients randomly probe a site to check whether there is still a server available. If not, d -- 1 other sites are probed and the task is assigned to the site with the fewest number of busy servers. If all the servers are also busy in each of these d -- 1 sites, the task is lost. This short paper analyzes a set of policies, i.e., (L, d) policies, that will occasionally probe additional sites even when there is still a server available at the site that was probed first. Using mean field methods, we show that these policies, that preventively probe other sites, can achieve the same loss probability while requiring a lower overall probe rate.