Hardware Acceleration in Ceph Distributed File System

Cloud computing becomes more and more popular providing the means to organize and deliver almost any kind of data and software services. Since it is inherently scalable to support rapid economic growth and productivity, its distributed data storage layer needs to be capable to address these requirements. A distributed file system contains a large number of cluster nodes and a large number of clients interacting with it. In this article, we propose two acceleration mechanisms based on multi-core network SoC to maximize each cluster node performance. First, the requests to the cluster node are balanced evenly on the platform cores, and second the requests are classified to decrease latency of sensitive operations and improve the overall cluster responsiveness and availability. For implementation, tests and measurements we have used a novel open source distributed file system: Ceph. Among many advantages over the competitors of Ceph we mention preservation of POSIX API, completely decoupled data and metadata and usage of object storage devices instead of block devices.