Abstract
Abstract
With the development of the information age, the database systems are expected
to provide real-time, accurate and high-performance services. In order to meet the
growing need of people, many large database applications have introduced distributed
in-memory object caching systems, of which Memcached is one of the most typical.
These systems put the most frequently accessed data in memory so that user requests
are processed without remote database operations. However, the traditional distributed
Memcached does not have the fault-tolerant capability. If one server node is crashed,
the lost data need to be reloaded from the remote servers or disks, which hurts the
system performance. In order to make the distributed Memcached more reliable and
available, Cocytus introduced Reed-Solomon codes and distributed protocol to the dis-
tributed Memcached which implement the fault-tolerant mechanism. Cocytus can saves
much memory compared to primary-backup replication when tolerating the same num-
ber of failures.
However, the relatively complex finite field calculation used by RS codes and the
high network transmission cost during data reconstruction are becoming a new perfor-
mance bottleneck of Memcached.
This paper introduced RDP codes into distributed Memcached to optimize the cal-
culation performance of Cocytus. In addition, this paper adopted RDOR codes and Col-
lective Reconstruction Read to speed up the rate of data recovery. The new distributed
Memcached has a similar performance with Cocytus in terms of data update and normal
data access. Compared with Cocytus which uses RS codes for fault-tolerant, the new
distributed Memcached with 4 data nodes and 2 check parity nodes reduces reconstruc-
tion time by up to 31%.
Key Words: Memcached; erasure codes; optimal recovery; parallel recovery
II
评论0