Asymptotically Optimal Load Balancing Topologies
Abstract
We consider a system of servers inter-connected by some underlying graph topology . Tasks arrive at the various servers as independent Poisson processes of rate . Each incoming task is irrevocably assigned to whichever server has the smallest number of tasks among the one where it appears and its neighbors in . Tasks have unit-mean exponential service times and leave the system upon service completion. The above model has been extensively investigated in the case is a clique. Since the servers are exchangeable in that case, the queue length process is quite tractable, and it has been proved that for any , the fraction of servers with two or more tasks vanishes in the limit as . For an arbitrary graph , the lack of exchangeability severely complicates the analysis, and the queue length process tends to be worse than for a clique. Accordingly, a graph is said to be -optimal or -optimal when the occupancy process on is equivalent to that on a clique on an -scale or -scale, respectively. We prove that if is an Erd\H{o}s-R\'enyi random graph with average degree , then it is with high probability -optimal and -optimal if and as , respectively. This demonstrates that optimality can be maintained at -scale and -scale while reducing the number of connections by nearly a factor and compared to a clique, provided the topology is suitably random. It is further shown that if contains bounded-degree nodes, then it cannot be -optimal. In addition, we establish that an arbitrary graph is -optimal when its minimum degree is , and may not be -optimal even when its minimum degree is for any .
Keywords
Cite
@article{arxiv.1707.05866,
title = {Asymptotically Optimal Load Balancing Topologies},
author = {Debankur Mukherjee and Sem C. Borst and Johan S. H. van Leeuwaarden},
journal= {arXiv preprint arXiv:1707.05866},
year = {2019}
}
Comments
A few relevant results from arXiv:1612.00723 are included for convenience