A Low-Latency Metadata Service for Geo-Distributed File Systems

Chuangwei Lin, Bowen Liu, Wei Zhou, Yueyue Xu, Xuyun Zhang, Wanchun Dou

Abstract: Geo-distributed file systems have been widely used by web services. An increasing number of time-critical web applications have been deployed on the cloud across geographical regions. In this circumstance, intolerant service latency will occur when user accesses remote servers. In view of this challenge, We design a metadata service which aims at reducing the service invocation latency. This low-latency metadata service is named LoLaMS. Taking advantage of the latency-aware dynamic subtree partition and migration, LoLaMS is capable to handle more metadata service invocations in the nearby metadata server. On account of that, the expected latency could be satisfied by LoLaMS. We implemented the LoLaMS and deployed it in a real-word cloud environment across different regions. The experimental results show that LoLaMS reduces the network latency effectively while ensuring high metadata consistency.