Distributed Object and Block Storages systems are studied in this thesis and their suitability as a storage solution for a dockerized cloud was evaluated. Docker is a relatively new virtualization framework. In beginning it was designed for containerizing processes on single host environments. However, it started to be used in multi host configurations and clouds, which has caused need for persistent storage solutions which are not relaying on host machine storage. Two open source distributed storage solutions were studied. Swift is an eventually consistent Object Storage system developed for the Openstack project. Ceph is a consistent storage system including object, block and file system storage subsystems. Swift and Ceph Object Storage systems were compared against each other. The Ceph Block Storage performance was evaluated against the virtual machine disk. The results show that Ceph has double the throughput in small objects from 8KB to 128KB compared to Swift throughput, and 30% better performance in files from 256KB to 100MB. The main trend between Swift and Ceph is that Ceph has better throughput on read operations in all object sizes. The Ceph Block Storage system was able to utilize 88.5% of the virtual machine disk write throughput.Throughput efficiency was calculated by multiplying write throughput of Ceph block by three and it dividing by virtual machine disk write throughput. Ceph block throughput needed to be tripled because replication tripled amount of disk writes. Ceph journal files were not stored on the disk so those wont affect efficiency.
展开▼
