2560 Switch, routing enabled, CPU 99%/3500pps/14Mb total bandwidth

[Thomas Rojsel]

ideas anyone?

Thanks,

/Thomas

[Arnaud_9]

First, I suppose you have swapped 5 and 6 in 2560, and you're talking about an HP 2650 ?

First of all, check in the log for any message which could be repeated many times
('sho log' command).

High CPU utilisation often occur with switching or routing loops.
Even if your statistics (3500pps/14Mb) are not really high for a switching loop, you should check following points :
- Is there any physical loop (Patch Cord between two ports, two link to another switch or hub) ?
- If there is one or more, is there an aggregation port and/or Spanning-tree enabled on the switch ?
- If spanning tree enabled and there is many switches in your network, try a 'show spanning-tree' in the CLI, and look at the Root Mac-address field. Is it the MAC address of your switch ?
--> Spanning tree can cause high CPU usage on the root switch in particular cases.


Look especially for a rooting loop :
- Check the routing tables of all other routers directly connected to your switch, and try to find loops (or opposite routes).
Begin with all next hops of the routing table of your switch.


Here are some other various questions :
- Your high CPU usage is constant, temporary, random ?
- Was the switch working properly before enabling routing ? What are new devices connected ?
- Where is located this switch in your network (Edge, Backbone, Servers connection, ...)
- How many Client/Server session should cross the switch in standard conditions ?
- Is the 'ip directed-broadcast' parameter enabled on the switch ? Is it indispensable ?
--> This can also cause burst in some cases.
Check the ratio of broadcast in the 'show interface' or 'sho stat' command.


Hope this will help you.
Arnaud

[Ron Kinner]

It appears that the 2650 is running off an initial release of code and there have not been any updates yet. So essentially you are running 1.0 code even tho they call it H.07.02 and I would not be surprised to find a software bug involved here. I see they already have discovered that if a telnet to the switch dies unexpectedly that you have to reboot the switch. Shouldn't be too long until they come out with a new version. I'd sign up for their automatic email notification if I were you.

If there is any chance of a loop try running spanning tree.

Make sure you haven't done some dumb like set the default route to point to the an interface on the switch. Verify that each of your static routes points to gateway which knows what to do with the packet and won't just send it right back to you. If any other switch or router points to your switch with static routes make sure that the switch knows what to do with the packets. Run a few traceroutes from hosts on the switch to see if there are any suspicious loops.

Ron

[Thomas Rojsel]

Thank you guys.

I got some response from HP tech-people and the says that i can't use this as a Layer 3 routing switch 'cause they only put in a HOST-buffer with 128 slots, so for every host connected over the 128 would consume cpu. If this is by design or a FW bug they would'nt say. I suspect i was a bit optimistic to use this one out in the internet. They recommended us to buy a 93XX. Well a little overkill for us though, will go with EXTREME SUMMIT24E3 'cause they're using wirespeed IP-routing for just a bit more money.

Thanks again.

/Thomas