I've got a packet loss problem I can't seem to figure out. On the network
in question, I've got several test scripts running. Each is doing a hundred
pings from a different point on the network every five minutes to another
host on the network (at least one hop away). No matter where I start from
or ping to, I see some amount of packet loss, perhaps 5-10 packets in an
hour. That's not a lot, I know, but I don't see why we should be seeing
any, except on rare occasions.
One segment of this network has two different subnets assigned to it, so
that traffic between the subnets has to go through the router at the common
FastEthernet interface. I know this isn't ideal, but it's part of a
workaround for another problem we can't seem to solve any other way.
Whenever a large amount of traffic (hundreds of megs) passes through the
router (from the FastEthernet port to an Ethernet port, or into the
FastEthernet and back out again), the packet loss shown by the ping scripts
increases dramatically, sometimes reaching as high as 12-15% in a single
set of 100 pings.
On the router that has the two different subnets out one FastEthernet port
(a 2621 with a 1-port Ethernet card), none of the interfaces are showing
any input or output drops. The 'show buffers' command shows some failures,
but these aren't increasing terribly much over time. There are quite a few
collisions on the Ethernet port, usually totalling about 6-9% of packets
output. In times of high traffic, throughput for other data streams is
noticably reduced. In the output of 'show process cpu,' the totals at the
top seem kind of high (to me at least), for example, five seconds: 38%/16%;
one minute: 22%; five minutes: 20%. The other figures in the list below
that don't really seem to change much from more idle times.
The problem actually was worse, and adding 'ip route-cache same-interface'
to the FastEthernet port made things a little better, but the problem is
still there.
Do we need a new router? Maybe more memory for the current one (it seems to
usually have 800k - 1.2mb free at any one time)? Do we need to tune the
buffer settings?
Thanks!
--Steve
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:08 EDT