[nsp] High CPU on 6400 NRP-1

[Clayton Zekelman]

I'm having a strange problem on a 6400 NRP-1.

With 20 users terminating with PPPoE, and CEF on, the router runs at around 
75% CPU, most of it interrupt.  If I turn off CEF, it drops to around 1% CPU.

I have another identical NRP that has around 450 users on it running at 22% 
CPU with CEF on.

If I do a "show alignment" on the first unit with CEF on, the Spurious 
Access counter climbs.  If I do it on the second unit, it does not.

Both are running 12.2(15)T1.  The configurations are roughly the same - the 
only difference is the one with problems has a MTU of 1460 on the Virtual 
Template, and the one without has an MTU of 1454.  I'm wondering if this is 
causing the problem.  The users have been bounced a whole bunch lately, so 
I'm not inclined to change it right now.

Interesting to note that I had a 7206 with an NPE-200 exhibiting the same 
strange behavior, but when I swapped out to an NPE-225, it went away - CEF 
worked fine.  From what I remember, the NRP-1 and the NPE-200 share some 
architectural similarities.

I also have a couple of NRP-1's terminating around 900 users each through 
L2TP running around 50% interrupt, with the Spurious Access counter 
climbing.  Doesn't seem to be huge issue right now.

I'm wondering if the MTU issue and fragmentation are related to the high 
interrupt load.  There is a bug ID (CSCeb87509) that talks about something 
similar, but we're not running SSG, and its listed as un-reproduceable.

I can flip CEF on, and watch the router load go up, then switch it off, and 
it goes right back down again.  I'm afraid that once I have more users on 
the unit, I'll need CEF to be able to handle the load.

In all cases, I have dot1q trunking on the routers - this was listed in the 
Bug ID as a condition.

Suggestions?  Insight?  Could it be fragging related to the MTU?



---
Clayton Zekelman
Managed Network Systems Inc. (MNSi)
344-300 Tecumseh Rd. E.
Windsor, Ontario
N8X 5E8

tel. 519-985-8410
fax. 519-258-3009