I'm seeing 2-5% packet loss going through a Cisco 2621 with <10mbps of traffic running at ~50% CPU. (packet loss based on ping results) Pinging another box on the same catalyst 2900 switch gives no packet loss, so it seems the 2621 is the source of the packet loss. I need help figuring out why it is dropping packets, and how to stop it. The odd thing is that the interface stats don't show any dropped packets: Full-duplex, 100Mb/s, 100BaseTX/FX ARP type: ARPA, ARP Timeout 00:01:00 Last input 00:00:00, output 00:00:00, output hang never Last clearing of "show interface" counters 00:10:16 Queueing strategy: fifo Output queue 0/40, 0 drops; input queue 1/512, 0 drops 5 minute input rate 5257000 bits/sec, 1383 packets/sec 5 minute output rate 5692000 bits/sec, 1448 packets/sec 845743 packets input, 392733148 bytes Received 403 broadcasts, 0 runts, 0 giants, 0 throttles 0 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored 0 watchdog, 0 multicast 0 input packets with dribble condition detected 887446 packets output, 430245866 bytes, 0 underruns 0 output errors, 0 collisions, 0 interface resets 0 babbles, 0 late collision, 22694 deferred 0 lost carrier, 0 no carrier 0 output buffer failures, 0 output buffers swapped out Ralph Doncaster principal, IStop.com
I'm seeing 2-5% packet loss going through a Cisco 2621 with <10mbps of traffic running at ~50% CPU. (packet loss based on ping results)
Isn't ping the first thing to be dropped in favor of other traffic? I remember a similar issue and Cisco saying that was the behavior. Don't quote me on that. jas
--- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu] On Behalf Of Jason Lewis Isn't ping the first thing to be dropped in favor of other traffic? I remember a similar issue and Cisco saying that was the behavior. Don't quote me on that. jas --- Even if it is, that still means that other packets could be lost had those pings not been there. --Phil
On Tue, 23 Jul 2002, Phil Rosenthal wrote:
--- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu] On Behalf Of Jason Lewis
Isn't ping the first thing to be dropped in favor of other traffic? I remember a similar issue and Cisco saying that was the behavior. Don't quote me on that.
jas ---
Even if it is, that still means that other packets could be lost had those pings not been there.
Not neccessarily. It's my experience that ciscos will sometime drop icmp instead of replying when under load...but that's only for packets directed at its interfaces. So, I might see 5% packetloss from the router itself, but 0% packetloss for everything behind it. Andy xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Andy Dills 301-682-9972 Xecunet, LLC www.xecu.net xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Dialup * Webhosting * E-Commerce * High-Speed Access
On Tue, 2002-07-23 at 10:53, Ralph Doncaster wrote:
I'm seeing 2-5% packet loss going through a Cisco 2621 with <10mbps of traffic running at ~50% CPU. (packet loss based on ping results) ...
If its not the switch or source ethernet segment generally, and it doesn't appear to be the router itself, I would concentrate on the ethernet segment between the switch and the router. (Assuming that your packet loss is while pinging the router, and not something on the other end of the router.) -Wes Bachman wbachman@leepfrogDOTcom -- Wes Bachman System & Network Administration, Software Development Leepfrog Technologies, Inc.
On Tue, Jul 23, 2002 at 11:53:29AM -0400, Ralph Doncaster wrote:
I'm seeing 2-5% packet loss going through a Cisco 2621 with <10mbps of traffic running at ~50% CPU. (packet loss based on ping results)
Pinging another box on the same catalyst 2900 switch gives no packet loss, so it seems the 2621 is the source of the packet loss. I need help figuring out why it is dropping packets,
Because pings are processed by the CPU which has already lots of better things to do than answering to pings. Answering ICMP is quite low-prio for routing engines. Trying pinging ERXes... you'll get "packet loss" (correct: "sometimes no answer") even though the RE twiddles thumbs. There is big difference between vendors in respect of ICMP echo reply priorization. Cisco is usually quite "good" at it (except for when BGP Scanner is running you'll see an increase of some dozends to hundreds of milliseconds in reply latency).
and how to stop it.
Get a faster router and/or just don't care that much about ping replies from routing engines of routers. :-) Packet loss while _forwarding_ packets counts, not when pinging loopbacks (yes, I know, there are many customers who can't tell the difference). Regards Daniel
participants (6)
-
Andy Dills -
Daniel Roesen -
Jason Lewis -
Phil Rosenthal -
Ralph Doncaster -
Wes Bachman