[OpenAFS] Vice Partition Effects on Disk I/O

Warren.Yenson@morganstanley.com Warren.Yenson@morganstanley.com
Fri, 7 Sep 2001 15:27:24 -0400 (EDT)


> i have no end of trouble with fast wide diff on suns.  what adapter
> are you running and could you check sunsolve to see if there are
> any patches for the adapter?  you can use prtconf to see what the
> driver has negotiated as far as sync speed.  try changing the cable
> as well.  any 'strange' messages appearing in /var/adm/messages?

We use QLogic cards. We have many servers with these cards and this type
of disk for other applications, and our benchmarks (on similar servers and
disks) have shown that this is a reliable solution. I can check individual
components, like the cabling, but not until we move the volumes off to
another server.

I was more asking what bottlenecks or algorithms could show themselves
with having larger volumes and larger numbers of files per partition.
(i.e. in a manner similar to the discussion about maximum sizes for
cache partitions).  We have over 40000 volumes on that server, or about
4500 on each partitions.

Since we use UFS, is there still that problem where new files or volumes
have an O(n) search in /vicepX before they can be created?

              - Warren