Greetings all:
I attempted to run a network backup of a 650 GB Oracle warehouse using NSR
5.2 across a FDDI network to a TL894 tape library (4 TZ89s) this weekend.
The performance that I had was absolutely terrible. I cut the parallelism
down to 2 for each tape drive, and I couldn't get any faster performance
than 700 KB/s per tape drive. The average write speed for all 4 tape drives
at the same time was about 2.1 MB/s. This warehouse is running on a dual
processor AS 8200 5/300 server with 2GB of RAM. The performance I was
getting was much less than any of my benchmarks that I had during my testing
period, however it was the first time I ran a NSR backup against this
machine.
I then started a backup against a dual processor AS4000 5/400 with 2GB of
RAM across the same network, with 250GB of data, and I was getting a write
performance of 1.0MB/s per tape drive, or an average of 4MB/s for all 4
drives, which is the performance that I expected to have. The parallelism
was set to the same as I had it for the AS8200. All I changed was which
client I was backing up.
Could the backup be bound by the client's processing power? I am running
the NSR server on a 4 processor AS8400 5/440, and it doesn't seem to be
memory or CPU bound, but the clients are when the backup is running on them.
Any ideas, tips, or shotguns would be appreciated,
Thanks in advance,
--Blake Roberts
blarob_at_hbsi.com
Received on Sun Dec 13 1998 - 19:34:32 NZDT