Really slow vos dumps to local disks

[Atro Tossavainen]

Norman,

> Well, I learned that the problem with slow "vos dumps" is apparently a well
> known (by everyone but me <g>) AFS/Rx issue.  Thanks to some clues dropped in
> one or two messages on the openafs-info list I modified my scripts to do "vos
> dump"s on each file server machine (instead of just the one), and of only those
> volumes that live on that particular file server (instead of -all- volumes).
>  The difference was like night and day.  I was able to dump ~5-6 GB of backup
> volume data per machine in anywhere between 15 - 30 minutes.  Unbelievable.

I've always only vos dumped on the server itself.

Out of interest, I decided to do a bit of "iostat"ing on today's dumps.

On our main server, what I seem to be getting is up to 8 MB/s. It's a
dual-processor (400 MHz) Sun E450, the /vicep* partitions are on RAID5
arrays behind Sun's RAID controller (DPT PM3755U2B) and I'm dumping to
/scratch which is on a single 18GB drive behind the system standard
SCSI controller. The /scratch is capable of taking in ~25 MB/s (writing
a gigabyte file from /dev/zero with dd) and the RAID array is capable
of giving out around 40 MB/s.

I then repeated the test with our new box. It's a single-processor (450
MHz) E220R, the /vicep* partitions are on RAID5 arrays on a Compaq
RA8000 fibre channel box, and the scratch partition is a DiskSuite
stripe of two partitions on the internal 18GB disks. The /scratch is
capable of taking in almost 30 MB/s. I'm seeing around 6 MB/s. Wonder
what gives? The RAID array is easily capable of giving out over 40
MB/s.

-- 
Atro Tossavainen (Mr.)               / The Institute of Biotechnology at
Systems Analyst, Techno-Amish &     / the University of Helsinki, Finland,
+358-9-19158939  UNIX Dinosaur     / employs me, but my opinions are my own.
< URL : http : / / www . iki . fi / atro . tossavainen / >