Howdy, all.


I'm wondering if any of you have war stories on the topic of tape
throughput debugging.  I've got some confusing things happening here,
and I'm looking for ideas.  Take this as a caveat for the
long-distance virtual volumes, if you swing that way. :)

This is all AIX, for starters.

I've measured single stream traffic to my remote site 350 miles away
at 60-80MB/s over and over again.  Recently, though, I've been seeing
several hour periods in which a given stream is limited to 15M/s.  Or,
it seems that way: 3 streams bobble around in the vicinity or 45M/s, 2
wander around in the vicinity of 30.

This is during times when the source machine and the destination
machine are both significantly idle, 50% or more.  So it's not a gross
'box is busy' thing.

It includes periods when I'm sending single large files on all the
streams, so it's not database contention for recording the ongoing
copies.

I can get straight network traffic (iperf) to fill the Gb connection
without a problem, so it's not a bottleneck in the intervening net.


Unfortunately, that leaves me with the Great Trackless Waste of TSM:
tape performance analysis.  Ugh.

I'm at the spot where I'm considering going Windows and just
restarting the things.  Ick.

- Allen S. Rout