We're running ADSM 3.1 on an RS/6000 R50 server (4x604e) with ~150 GB of
SSA DASD, of which ~60 GB is being backed up daily.  We're using a 3590
drive as our backup device.  Oh, and it has 2 GB of RAM.


These are some older statistics (when we were backing up less...)

Total number of bytes transferred:  36.02 GB
Data transfer time:                 5,616.98 sec
Network data transfer rate:         6,724.53 KB/sec
Aggregate data transfer rate:       1,348.57 KB/sec

How I am interpreting this is that over the network (TCP/IP, 100BaseT
Ethernet) I'm getting ~6.5 MB/s (around 65% utilization) which is
acceptable.  However, the aggregate (as I'm interpreting it) would be the
overall throughput of the system.  Now, according to the 3590
documentation, it should be able to back up at up to 9 MB/s.  The SCSI-2
bus driving the 3590 can handle at least SCSI-1 speeds of 10 MB/s.   The
SSA driving our DASD can handle well over SCSI-2 speeds; so the drive and
network I/O don't seem to be the limiting factor.

What I'm wondering if there is not enough iron behind the ADSM server,
which is causing the performance problem - that is, why we can only back up
1.3 MB/s when all of the subsystems can handle more.  We've tried all kinds
of performance optimizations on our TCP/IP settings, as well as using over
512 MB of RAM for ADSM cache, with minimal performance improvement.  We've
tried the stuff in the ADSM Performance Tuning Guide (v3) and the folks
from IBM don't seem to have any clear-cut answers.

What I'd like to know is if a) anyone else has had similar types of
performance problems and b) what is a decent typical backup speed?  Am I
being too ambitious in wanting ~9MB/s?

--
Matthew Emmerton, System Administration
Matthew Emmerton, System Administration
Agora Food Merchants, National IT
+1 (905) 565-4231