[plug] Nick are you there

[raven at themaw.net]

Our discussion about filesystem performance on Monday night has got me 
thinking.

You mentioned that Linux filesystem performance is better that Solaris UFS 
and Veritas vxfs. I guess we are talking about ext[2,3], any others.

Do you think file size limitations in the ext file systems could be 
contributing to this?

What is the size of the differences?

Another thing that this brings to mind is that it implies the Linux 
VFS layer is also efficient and is unlikely to be a performance bottle 
neck. Solaris NFS outperforms Linux NFS for straight volume read/write by 
at least 20% in test results I have seen. My conclusion is that there must 
be some serious latency in one or more of NFS, sunrpc, IP stack or the 
network driver subsystems. Any thoughts on which of these might not be a 
source of latency?

Ian