[c-nsp] Constant output drops on etherchannel

Nick Hilliard nick at foobar.org
Sun Jan 16 10:14:39 EST 2011


On 16/01/2011 02:30, Dan Letkeman wrote:
> Drops are happening even when its not under load.  Has nothing to do
> with bandwidth.

Dan,

hypothetically on a 100Mb port, if you burst your output to 200 megs for 1 
second, then drop to zero traffic for 4 minutes 59 seconds, you will see:

- 50% packet loss on the link
- a 5 minute throughput rate of 333000 bits per sec

This is called a microburst.  I.e. a burst of traffic which goes beyond the 
capacity of the link, but which is too short to be measured accurately by 
your 5 minute rolling average.  Typically you'll see this on slower speed 
lan links with bursty traffic, and it's why you're seeing relatively low 
levels of traffic, but output drops on the interface.

If you want to fix this problem, you have several potential workarounds:

- increase your port speeds
- get a switch with bigger buffers
- tune the output buffers on your existing switch
- in your particular case, you could try fiddling with the etherchannel 
hashing algorithm to see if it helps (it's unlikely to make the problem 
disappear completely).

Going back to your port channel

> Port-channel2 is up, line protocol is up (connected)
>   Hardware is EtherChannel, address is 001b.d59d.7199 (bia 001b.d59d.7199)
>   MTU 1500 bytes, BW 200000 Kbit, DLY 100 usec,
>      reliability 255/255, txload 24/255, rxload 2/255
>   Encapsulation ARPA, loopback not set
>   Keepalive set (10 sec)
>   Full-duplex, 100Mb/s, link type is auto, media type is unknown
>   input flow-control is off, output flow-control is unsupported
>   Members in this channel: Fa0/23 Fa0/24

Your problem is here ------> ^^^^^^^^^^^^^

You need to upgrade your switch to a gig capable device.  You've outgrown 
your existing equipment.

Nick


More information about the cisco-nsp mailing list