Next: Experimental Results on Write
Up: Duplication Protocols
Previous: Protocol 4: Synchronous Client
In these protocols, the file system caches on the servers are
fully utilized to improve the overall I/O performance. Thus there
is a possibility, however small, that data can be lost when the
disk or the server node crashes before the cache data is written
onto the disks. Nevertheless, if a ``hard'' reliable storage
system is required, we can potentially use techniques such as
forced disk writes in these four duplication protocols. While the
four protocols with forced disk writes improve the reliability,
the penalty on the I/O performance are too heavy to make the
forced disk writes appealing. In addition, even if the forced disk
writes are used, these four duplication protocols still present
different performance and reliability.
Yifeng Zhu
2003-10-16