- 论坛徽章:
- 0
|
I suggest you learn them one by one.
1. no. pvfs, lustre, coda all have specific layer to deal with I/O. They are very different from NFS. The function layer you see is exactly not "virtual" but real. They are Parallel File System with latest network(GibE/IB/Myrinet) Distributed technology.(this is the reason why we like to call them Distributed Filesystem as well).
2. yes. pvfs2/lustre has low level I/O implementation. And this is the key factor when we measure their performance and think of tuning on them.
3. no. gfs has lock(DLM), membership mgm(provided by cluster mgr inside the rhcs) and other parts as standard cluster file system do has. Not only gfs but lustre can resident on a single server and export it's storage space to other consumers.
[ 本帖最后由 nntp 于 2006-9-28 00:14 编辑 ] |
|