武汉邮电科学研究院硕士学位论文
II
Abstract
Ceph is a decentralized open source distributed storage system. Because of its
unified interface, good scalability and availability, it is widely used in production
practice. Because of Ceph’s decentralized architecture, all storage nodes need to store
and manage metadata. So there is going to be a lot of small random chunks of data to
read or write. In addition, Ceph shred and hashes the data block as it processes read or
write request. As a result, the consistency and size of data blocks are reduced.
Considering these feature of Ceph, if the storage all use only HDD, it will greatly
limit the performance of Ceph. Although SSD can meet the needs of Ceph, their cost
is too high. Therefore, it is not suitable to use SSD to store all data in Ceph. Designing
a hybrid plan based on SSD and HDD is important to optimize Ceph’s performance.
Therefore, this paper studies the read-write performance of Ceph from the
perspective of hybrid storage, and achieves the following results:
(1) A complete hybrid storage optimization scheme is designed and implemented.
Experiments show that this scheme improves the IOPS and throughput of the virtual
hard disk which created by Ceph greatly.
(2) In order to solve the algorithm defect of Cache Tier, An elimination
algorithm based on gaussian distribution is designed. Simulation results show that this
algorithm performs better than LRU and LRU-2 when user access conforms to a
gaussian distribution.
Key words: Ceph; Distributed storage system; Hybrid storage; Cache Tier;
Elimination algorithm
评论0