BibTeX
@MISC{Feng_oht:hierarchical,
author = {Kun Feng and Tianyang Che and Tonglin Li and Ioan Raicu},
title = {OHT: Hierarchical Distributed Hash Tables},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
This paper presents OHT, a hierarchical distributed hash table, which improves the performance of practical ZHT. When n-to-n connection is needed, every single node has to keep a large number of socket connections. This might be quite expensive when n is extremely large. By performing a hierarchical distributed hash table, OHT can solve this problem. Theoretically, we are able to support an OHT system who consists of millions of nodes. This paper is a final report to give a problem statement, describe related work, and illustrate our design idea and implementation details of OHT. We have evaluated OHT’s performance under a Linux cluster with 512-cores. We scale OHT up to 32 physical nodes. The results shows that latency of OHT is around 2.7 ms and throughput can reach 12K ops/s for a 32-node scale. The experiments also show that OHT can work correctly even when failures arise. At the cost of losing some percentage of performance, OHT can be much more scalable.
