2610 switches losing management connectivity during higher peaks of IGMP and MLD traffic

[mjsaarin]

Hello,

Has anyone else seen 2610 switches losing management connectivity if there is more IGMP or MLD traffic passing through the switch? We have seen this few times and the only models affected are 2610s. The switches stop responding or respond very slowly to SNMP reequests or telnet/SSH login attempts. We have several other ProCurves in our net and they don't suffer from this issue. Also, the issue can be observed with different software versions on 2610 so I'm beginning to suspect that this has something to do with how the hardware handles multicast. The hosts sending the multicast are usually Windows boxes trying to reach DNS servers via multicast DNS (a service which we don't provide).

Cheers,

Matti

[TW_Nighty]

Hey Matti,

Check the cpu load of the affected switches.

and also check the latency to the affected switches using a basic ping command to the management interface (IP).

I have seen 2610 switches stop responding, with cpu loads of around 60%

You might find your switches also suffer from a high CPU load due to multicast or other traffic that needs passing trough software.

here's a show cpu from a random 2610 currently in production:

 show cpu
3 percent busy, from 3440307 sec ago
1 sec ave: 8 percent busy
5 sec ave: 15 percent busy
1 min ave: 6 percent busy

[mjsaarin]

High CPU could be the reason. Unfortunately, the switches become so unresponsive that it is not possible to issue any show commands. Ohterwise the switches work fine: spanning tree doesn't flap and traffic flows normally through them. Probably, I will need to build a minimal setup, begin sending IGMP packets and monitor the switch behaviour at the same time.

[TW_Nighty]

If possible, you can try a console connection.

Even under high stress the console worked for me, this does mean you need to be at the switch of course.

I was able to do some troubleshooting this way.