Nathan Neulinger writes:
>> We use IBM ADSM backup client which can backup and archive AFS/DFS at file
>> level, as well as ACLs and mount points. The binaries is different from ADS
>M
>> unix client -- they are dsmafs, dsmdfs. We developed our own wrapper for thi
>s
>> adsm binary so that 40, 000 AFS volumes, millions of files can be backed up
>> within reasonable time frame.
>
>I'd be interested in hearing about what you did to improve performance,
>as we've recently been forced to switch to using Seagate BkupExec to
>back up our unix machines. I've got a home grown libc wrapper that fakes
>out the open and read system calls to use vos dumps instead, however
>this is really really slow. Even on file servers with disks that are
>known to do 25 MB/sec sustained to the host, and 100 Mbit connection
>through a single switch, we still only get 3-4 MB/sec vos dump rate.
>
>Because of this, we've considered doing file level backups. What'all did
>you do to make the backup complete quicker?
>
>-- Nathan
Here is some information regarding Stanford ADSM configuration:
Two backup servers running on AIX 4.3.2. Each connected to two ethernet
interfaces and 4 tape drives in an IBM tape library (LIB3575). The servers are
configured with 50M EMC disks as backup cache so the backup will go to the
disks first and flushed to tapes at later time. The backup wrapper around the
backup program will run 10 backup processes (configurable) concurrently. The
system backs up AFS volumes at file level, plus Stanford campus mail servers.
The backup takes about 4-5 hours for about 20GB (daily changes) on each
machine. The network transfer speed averaged 10M/sec.
Hope this helps,
Susan
