Constructing a Two-Level Topology-Aware Distributed Hash Table with a Hierarchical Network Coordinate System

Abstract

Distributed Hash Tables (DHTs) may suffer from a high lookup latency due to the issue of topology mismatching between the overlay and the physical underlying network. To handle this problem, recently, some researchers use network coordinates (NCs) to construct topology-aware DHTs(e.g. Canary). However, these proposed DHTs still have room to be improved due to the inaccuracy of NCs. To construct a more efficient DHT, in this paper, we propose hCanary: a two-level topology-aware CAN. The novelty of hCanary is that it uses a hierarchical NC system-Pharos to generate more accurate NCs to construct multiple local topology-aware overlays over a global topology-aware CAN to reduce lookup latency. Meanwhile, we further use expressway technique to improve the ability of query forwarding of hCanary. Experiments show local overlays of hCanary can reduce average lookup latency by up to 60% and hCanary with expressways can reduce average lookup latency by up to 77% compared with Canary.