Randomized load balancing by joining and splitting bins

Authors:
James Aspnes;Yitong Yin
Affiliations:
Department of Computer Science, Yale University, United States;State Key Laboratory for Novel Software Technology, Nanjing University, China
Venue:
Information Processing Letters
Year:
2012

Citing 3
Cited 0

Consistent hashing and random trees: distributed caching protocols for relieving hot spots on the World Wide Web

STOC '97 Proceedings of the twenty-ninth annual ACM symposium on Theory of computing
Uniform Data Sampling from a Peer-to-Peer Network

ICDCS '07 Proceedings of the 27th International Conference on Distributed Computing Systems
On unbiased sampling for unstructured peer-to-peer networks

IEEE/ACM Transactions on Networking (TON)

Quantified Score

Hi-index	0.89

Visualization

Abstract

We analyze the performance of a very natural randomized load balancing scheme: uniformly joining and splitting weighted bins. We develop a norm-based technique for analyzing this simple procedure. By applying the technique, we prove several bounds for the expected load factor. Specifically, if we keep uniformly splitting the bins without joining them, the expected load factor is between @W(n^0^.^5) and O(n^0^.^7^4^2), however, if we alternatively join and split bins, the expected load factor converges to O(n^1^/^1^2^l^o^g^"^2^n). These bounds justify the intuition that the power of being adaptive to the current loads is essential for load balancing tasks, and they also show a somehow surprising phenomenon that joins can actually help load balancing if such power is not available.