There may be more to this story, but this caught me with such surprise
that I thought it might be worth sharing. I don't know much about tuning
TCP settings, except that it is not an exact science. I have to wonder
if someone should have seen this problem coming: this 996/200 has a gig
of memory, and the "Inbound Buffer Pool" (actually the Maximum Inbound
Buffer Memory in NETXPORT.GLOBAL) is at the default, minimum value. Now
that I've said it, it sounds like a bad thing. I am recommending that we
revisit these, most obviously Number of Outbound Buffers in
NETXPORT.NI.LAN2. Unless I wrong about that...

Can anyone make some recommendations for how we should tune these
settings, for this configuration? We ftp send and receive 50+ MB files,
and use telnet heavily. Fortunately, we have a 968/RX with 512MB to try
these changes on, and see how they go, but I expect any results of
testing there will only provide general guidelines, not hard answers.

BTW, the results of this change are that our CONNECTION FAILURE DETECTED
(SOCKERR 67)s have stopped happening, and successful transfers take
about half the time that they were when they worked. We still see REMOTE
ABORTED THE CONNECTION (SOCKER 64)s, at the interval of our Connection
Assurance Timeouts. I believe this to have to do with the WAN we are
going across, but that leads to my next question: Our Network Segment
Size (Bytes) in NETXPORT.NI.LAN2 is at 1984, but the largest packet I
can ping the remote ftp server with is 1024. I would think that I would
not want my Network Segment Size to be above that, right? Or, do I need
to persuade the keepers of the WAN that 1024 just isn't big enough?

> -----Original Message-----
<snip>
>      The FTP jobs did indeed hang again last night and per our action
> plan
>      the staff immediately called the HPRC and a Senior Eng. logged on
>
>      and looked around.  What he found was that we had the Inbound
> Buffer
>      Pool for LAN2 (Token Ring) filled up.  The default and minimum
> value
>      for that entry is 256 (which is what it was set at).  Taking into
>
>      consideration the amount of memory this box has on it (1Gb) the
> RC
>      Engineer suggested this value be upped to 2048.  It was and the
>      network and your jobs were restarted.  The RCE stayed online and
>      watched them successfully complete.
>
>      These entries were checked early on (by myself and another
> engineer)
>      but since the Network had already been bounced (brought down and
>      attempted to be brought back up) we didn't see this table being
> filled
>      (thus the action plan to be able to troubleshoot the problem in
> real
>      time as the problem was occuring and no immediate pressure to get
> the
>      network back up).
>