Cell Relay Archive

[dave bernardi <bernardi@kennametal.com>]

I had a wierd problem where performance on certain connections through a
Fore ASX200 would slow to a crawl and not knowing what else to do I
rebooted it and all was well. I'd rather determine what the problem is and
not get into the habit of rebooting.

The edge devices are Bay Centillions (with Ethernet PC's behind it),
multihomed via OC-3 to 2 Fore switches, each Fore switch has a couple of
OC-12 links to other Fore switches with more Bay Centillions at the other
end of the cloud. 

LANE is done in the Bay(s). Fore is simply setting up SVC's. Routing tables
are static.

I got a call stating that performance between a group of PC's on one edge
was experiencing terrible performance talking to a server on another edge
(same VLAN/ELAN). The sniffer (on Ethernet) showed an average response time
of 700ms with no retransmissions, lost packets (implying no cell loss on
the ATM backbone), just real slow.

When I pulled one of the OC-3 links performance went back down to ~5ms
(after 10 or 20 seconds). Plugged the OC-3 link back in, restarted the the
Centillion, reestablished the PC/Server connection, same thing. 

The problem appeared to be on the Fore switch (the one at the other end of
the pulled OC-3 link). I tried all the obvious show commands but nothing
looked out of the ordinary (compared to the other switches). I looked at
bandwidth on the port(s) in question and none were over 20K, no lost cells,
no overflows, no errors. There were about 200 SVC's built and there were no
problems or timeouts with call setups, just performance. 

One thing that did seem strange to me was that were SVC's that have been up
for over 120 days but had 0 Cells Received. Shouldn't the switch tear down
SVC's that aren't being used? 

Maybe this is just a wierd Fore problem and I just need to upgrade but I
hate to just blindly do that. Current software version is FT410.76 and the
box has been up over 140 days if that helps. All 5 other Fore switches in
the backbone have not exhibited this problem, yet...

Thanks in advance!




--
Dave Bernardi
bernardi@kennametal.com