On 7/28/2011 8:30 PM, Mark Ranft wrote:
> While setting timers is important. It doesn't explain the deterioration.  My bet is the switch port setting is not 100 full. This would explain why the network would still work. And there is a slow down.  I also bet pings would drop packets. 

With mismatched duplex, both sides will see errors, just different
types.  The Linkcontrol stats are relatively clean, other than received
address drops.  The percentage of broadcast packets is rather large
relative to unicast, but maybe this is just a busy segment with
relatively little 3000-destined traffic (?).  Since the 3000 side is
100/full, I would expect CRC errors or similar (TCP retransmits, which
Linkcontrol won't tell) when the switch side "detects a collision" and
stops transmitting and/or retransmits.

Were you doing pings from the same subnet, or a different one?

You have two active network interfaces.  What does your routing look
like?  Are your gateways as expected (you may have had redirects, which
the 3000 honors, but never forgets).  You may have some asymmetric
routing taking place (received traffic on one interface being answered
on another).  Or iss one of the interfaces "strictly" a local subnet
with no gateway?

If pings are that unreliable, there is packet loss somewhere, but it
doesn't appear to be on the 3000-to-upstream hop.  Once you get packet
loss, then the default timers can really exaggerate the delays.  Fixing
the timers may reduce the delays, but won't eliminate the packet loss.

Jeff

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *