MLB-Wrapper: Distributed High Scalable BP-Wrapper

Abstract

In distributed system, concurrent accesses lead to lock contention which greatly increases response time. Cache can reduce response time, but cannot gain scalability. Currently, the common solution is to modify the caching algorithm itself, although the scalability can be improved, the hit ratio hurt. In this paper, we propose a system framework, called MLB-Wrapper, which resolves the scalability issue in distributed systems. The framework does not modify the cache algorithm itself. Therefore, it preserves high cache hit ratio about the replacement algorithms. Multi-levels distributed cache is used in MLB-Wrapper which adopts batching technique in every level to eliminate the lock contention. Furthermore, MLB-Wrapper is flexible because different replacement algorithms can be used in each level. We describe the main idea and the detailed implementation of MLB-Wrapper in this paper. Experiment results show that, in the intensive concurrent environment, MLB-Wrapper can not only reduce lock contention but also have higher scalability.