[c-nsp] Best practice - Core vs Access Router

Andy B. globichen at gmail.com
Tue Feb 9 08:20:06 EST 2010

On Tue, Feb 9, 2010 at 2:04 PM, Saku Ytti <saku at ytti.fi> wrote:
> Could you try to catch this when the five second value is >40% so we'll see
> what is causing the load. Currently what ever is happening, is not
> happening.

Actually, last time when the core started to become very unresponsive,
CPU load was lower than usual - 12-15%. Except when it was
re-establishing BGP sessions with transit customers, then it went up
to 99% for quite a while, but that is normal.

> Output 5s sorted list.
>>                      Module       pps   peak-pps                     peak-time
>>                      5        4443376   10849623  12:44:28 CEST Mon Dec 21 2009
> 10Mpps peak, well within limits of CFC system, so you're not anywhere near
> the performance limits.
>> Netflow Resources
>>           TCAM utilization:       Module       Created      Failed       %Used
>>                                   5             262020           0        100%
> Netflow full, highly typical and nothing to worry about.
>> I do see input drops - what does that mean?
> It means that you got more packets towards software than buffers could
> hold, default is 75 packets, which is way too little for even some normal
> situations, such as BGP, especially route reflector use.
> If it is normal, you should increase it to 1k or 2k.
> But it might also indicate that transit traffic is coming to control-plane,
> there are many tools 7600 offers to troubleshoot them.  When you look at
> those interfaces where you see drops, do any of them display packets in
> input buffer /right now/, if so, you can use 'show buffers input-interface
> X header' to see those packets are, which will go long way to determine if
> they are normal or something to worry about.

Yes, here is some input buffer:

#show buffers input-interface te9/1 header

Buffer information for Small buffer at 0x5007A3A8
  data_area 0x806EBC4, refcount 1, next 0x454585E0, flags 0x280
  linktype 7 (IP), enctype 1 (ARPA), encsize 14, rxtype 1
  if_input 0x4646B618 (TenGigabitEthernet9/1), if_output 0x0 (None)
  inputtime 47w4d (elapsed 00:00:47.772)
  outputtime 47w4d (elapsed 00:00:30.952), oqnumber 65535
  datagramstart 0x806EC3A, datagramsize 62, maximum size 308
  mac_start 0x806EC3A, addr_start 0x806EC3A, info_start 0x0
  network_start 0x806EC48, transport_start 0x806EC5C, caller_pc 0x4187C1F0

  source: x.x.72.173, destination: y.y.161.0, id: 0x611D, ttl: 120,
  TOS: 0 prot: 6, source port 60922, destination port 47743

#show buffers input-interface te9/2 header

Buffer information for Small buffer at 0x5002EA70
  data_area 0x802C2C4, refcount 1, next 0x5007DA28, flags 0x200
  linktype 7 (IP), enctype 1 (ARPA), encsize 14, rxtype 1
  if_input 0x4646CC78 (TenGigabitEthernet9/2), if_output 0x0 (None)
  inputtime 47w4d (elapsed 00:00:54.296)
  outputtime 47w4d (elapsed 00:00:43.944), oqnumber 65535
  datagramstart 0x802C33A, datagramsize 70, maximum size 308
  mac_start 0x802C33A, addr_start 0x802C33A, info_start 0x0
  network_start 0x802C348, transport_start 0x802C35C, caller_pc 0x4187C1F0

  source: y.y.226.89, destination: x.x.160.112, id: 0x5ADD, ttl: 52,
  TOS: 0 prot: 6, source port 52067, destination port 18309

Changes all the time. Sometimes it is empty, but it seems rarely to be the case.

More information about the cisco-nsp mailing list