Since there seems to be a certain amount of interest on the list,
here's a list of things we have (and have not :) tried to date. First,
a quick rundown on the filesystem:

  * 522 GB hardware RAID partition.
  * Approximately 50 GB in 2,995,971 files (average of 16 kB/file)
  * Fibre channel 1 Gbps interface.

General consensus was that we are being hit by the "many small files"
problem, which certainly fits in with our situation.

Things we've tried to date:

  * Increased the DNLC and inode cache (see
    http://www.princeton.edu/~unix/Solaris/troubleshoot/kerntune.html --
    very useful web page)

    Slight improvement from this, but Networker is still timing out.

  * Running uasm over the whole filesystem, piped to /dev/null to
    eliminate the network as a bottleneck -- I aborted this after
    eight hours (considering that we're "only" using 60 GB of space at
    peak, and less than 50 GB after the processes using the disk clean
    up after themselves, eight hours _should_ be adequate.)

  * Removing the temporary working area from the backup -- again, maybe
    a slight improvement, but it doesn't resolve the underlying issue.

Executive summary: we've managed to get slight improvements, but Networker
still likes to time out whilst backing up this filesystem. Sigh. Too many
small files.

At the moment, we're preparing to evaluate Sun's QFS as a possible
alternative to UFS for this particular situation. Does anybody have any
knowledge of how Networker interacts with QFS? Sun's representative is
saying that our situation seems ideal for QFS, but then, I'd be inclined
to take his words with a grain of salt :-) The biggest problem from my
point of view is that Legato's compatibility matrix only lists UFS and
VxFS; VxFS is an option, but it would be nice (if only from a bargaining
point of view ;-) to have QFS as an option as well.

As an aside, I'll be going on leave at the end of this week for a six
week holiday, so the testing will be somewhat abbreviated (we have a
thirty day evaluation license) ... I'll be picking up the ball once I get
back, in late October.

Ta muchly,

Stuart.

--
Note: To sign off this list, send a "signoff" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=