Thanks Lisa and Richard.

The assertion by our Network group that the communications pipe
is ok is based on the packet trace using a sniffer.
The frame size for most packets it 1514, according to the sniffer
trace.  The network card is set to auto negotiate and I'll check
on that again.  However, our network engineer used the
millisecond response time between packets in the trace to
determine the actual throughput.  When we are sending data and
the windows size appears to be 63K, he feels we are sending a
constant 5MB/Sec.  The throughput goes much lower when it appears
the NT client machine starts putting the brakes on and the
TCPwindowsize shrinks, sometimes to zero.  Why it puts the brakes
on is the mystery. Here are the stats from the restore mentioned
in my previous post:

ANS1896I ***** Restored 74050 objects *****
ANS1896I ***** Restored 74100 objects *****
ANS1896I ***** Restored 74150 objects *****
ANS1896I ***** Restored 74200 objects *****
ANS1896I ***** Restored 74250 objects *****
ANS1896I ***** Restored 74300 objects *****
ANS1896I ***** Restored 74350 objects *****

Total number of objects restored:    74,390
Total number of objects failed:           3
Total number of bytes transferred:     4.88 GB
Data transfer time:                15,169.42 sec
Network data transfer rate:          337.96 KB/sec
Aggregate data transfer rate:        233.96 KB/sec
Elapsed processing time:           06:05:12
tsm>


I should also mention that client compression is off for this
client, collocation is turned for the tape storage pool involved
and not many tape mounts were required.  We use 3590 E1A tape
drives.

I don't have a clue. Anyone else?

I have opened a problem with TSM support and am waiting for a
call back from the TSM performance team.

Thanks,
Jeff Connor





Richard Sims <rbs AT BU DOT EDU>@VM.MARIST.EDU> on 09/13/2000 03:34:24
PM

Please respond to "ADSM: Dist Stor Manager"
      <ADSM-L AT VM.MARIST DOT EDU>

Sent by:  "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>


To:   ADSM-L AT VM.MARIST DOT EDU
cc:

Subject:  Re: Slow restore for large NT client.. help!


Jeff - Lisa's insightful response made me think of something...

You say you are using 100Mbit Ethernet.  A gotcha in that area to
make sure of is that your adapter cards are not set for
Auto Negotiation: it is notorious for not working.  Too often,
one ends up with a conflict in duplex settings that results in
substantially reduced throughput, with Late Collisions showing up
in the ethernet adapter stats.  Your network staff "feels the
bandwidth is there", but that's theoretical and means little.
Unless all components of the network agree in their
configuration,
throughput is hampered.  Just another thing to look at.
  Richard Sims, BU