서강대학교

초록/요약

With the proliferation of cloud computing technologies, the Ceph, a distributed objected-based storage system has been an attractive alternative to building a storage backend due to its excellent performance, reliability, and scalability. As the storage system processes huge amount data and the network traffic generated from the cloud increases rapidly, designing a high-performance messenger in the storage system has created a lot of challenging issues. One of Ceph's messengers, async messenger, is known for its efficient and flexible communication structure. However, to solve Ceph's network bottleneck, performance improvement of async messenger is essential. The async messenger's worker threads typically handle multiple connections at the same time, so there is a limit to handling multiple connections serially. It also schedules communication threads using a simple round-robin technique without considering the various workloads generated from multiple connections. This causes a load imbalance between the communication threads, degrading the performance of the Ceph storage system. This paper parallelizes the message reception process to use the messenger 's worker thread efficiently. Also, to solve the problem of load imbalance, this paper proposes a scheduling method using genetic algorithm. Finally, this paper improved the performance of the async messenger by combining the load balancing scheduling method and the parallel structure. As the result, async-read2-GA have 8∼20% higher performance than the existing async messenger.