Achieving Better Load Balance in Distributed Storage System

摘要

The CPU processing speed and disk capacity are increasing tremendously during the past decade. However, the even faster increasing number of users generates higher requirements for high performance and huge capacity computer systems. More and more applications are now running on distributed systems. Load balance is an important issue in distributed storage systems. Currently, very few of them has a finely-tuned load balance scheme to achieve higher system throughput and shorter client response times. We propose a new load balance scheme called Cooperative Load Balance Scheme (CLBS) to solve this problem. In CLBS, servers cooperate with a client in its decision of the next request by sending their current workload information to the client every small amount of time. With the assistance of servers, a client can always choose the light-loaded server. We did trace driven simulation experiments to compare CLBS with the scheme used in a traditional distributed storage system. The results show a shorter average response time and a smaller average queue size under both light and heavy workloads.