Switches with High CPU Load

[jorge.s]

Hi,

I've some switches, mode Catalyst 3500XL with really high CPU load, unjustified.

Would you know how can I identify what is creating this CPU Load?

here is the result of sh proc:

EDE0002#sh proc

CPU utilization for five seconds: 59%/10%; one minute: 58%; five minutes: 58%

PID QTy PC Runtime (ms) Invoked uSecs Stacks TTY Process

1 Csp 1E8BCC 8429 238252 35 2648/3000 0 Load Meter

3 Lst 1D46B4 794238 141004 5632 5736/6000 0 Check heaps

4 Cwe 1CC354 0 1 0 5604/6000 0 Chunk Manager

5 Cwe 1D99CC 419 651 643 5644/6000 0 Pool Manager

6 Mst 15D3F8 3 2 1500 5600/6000 0 Timers

7 Lwe A7238 29 9 3222 5640/6000 0 Entity MIB API

8 Mwe 247C2C 40076 295951 135 5468/6000 0 HC Counter Timer

9 Lwe 25C6E4 2708049 5404800 501 4940/6000 0 ARP Input

10 Hwe EB27C 0 2 0 2752/3000 0 RAM Access (dm 0

11 Cwe 1DED9C 0 1 0 5640/6000 0 Critical Bkgnd

12 Mwe 1AB4B8 75616 445039 16910652/12000 0 Net Background

13 Lwe 1531B8 550 1574 34910560/12000 0 Logger

14 Msp 16FF64 236848 1161103 203 5540/6000 0 TTY Background

15 Msp 1AAB1C 5891844 5766873 1021 5056/6000 0 Per-Second Jobs

16 Hwe 1AAD2C 221397 2533012 87 5632/6000 0 Net Input

17 Csp 1B2CB8 24529 238252 102 5636/6000 0 Compute load avg

18 Msp 1AAB68 292213 19855 14717 5624/6000 0 Per-minute Jobs

19 Hrd EA844 62712921 121567040 515 2120/3000 0 LED Control Proc

20 Msp 112CE4 8160516 1161104 7028 1244/3000 0 Frank Aging

21 Hsp 212224 308215965 217439889 1417 3464/6000 0 Port Status Proc

22 Mwe 23CA20 231 1566 147 4076/6000 0 VM Prune Events

23 Mwe 3349E0 0 1 0 5796/6000 0 IP NAT Ager

24 Msp E95F4 10160039 2265016 4485 2516/3000 0 GDS Frame Ager

25 Hwe EB27C 0 1 0 2784/3000 0 RAM Access (gi0/

26 M* 0 35 36 97210112/12000 1 Virtual Exec

27 Hwe 114640 0 1 0 2752/3000 0 Frank Global Mas

30 Hwe EB27C 0 1 0 2788/3000 0 RAM Access (gi0/

32 Mrd EA844 47914828 72466461 661 5152/6000 0 Broadcast Storm

33 Hwe 213C78 0 1 0 2644/3000 0 Port Group Chang

35 Hwe 226784 1412849 450546 3135 1332/3000 0 Address Learning

36 Hwe EB27C 0 1 0 2784/3000 0 RAM Access (dm 1

37 Hwe EB27C 0 1 0 2784/3000 0 RAM Access (dm 2

38 Hwe EB27C 0 1 0 2784/3000 0 RAM Access (dm 3

39 Hwe EB27C 0 1 0 2788/3000 0 RAM Access (dm 4

40 Hwe EB27C 0 1 0 2788/3000 0 RAM Access (dm 5

41 Msp 104F70 46290393 8783406 5270 5512/6000 0 Enet Aging

42 Mwe 259EC 408602 2334358 17510204/12000 0 IP Input

PID QTy PC Runtime (ms) Invoked uSecs Stacks TTY Process

43 Lwe 22C7DC 11102061 321491 34533 1956/3000 0 Address Deletion

44 Msp 226714 29085 239717 121 2624/3000 0 Address Sorting

[Georg Pauwen · Accepted solution]

Hello Jorge,

actually, this is default behaviour for the 3500XL (and 2900XL) switches, the CPU utilization is always between 30 and 50 percent, even under minimal load.

Have a look at the document below, it explains the default CPU utilization:

Catalyst 2900XL/3500XL Switches

Baseline CPU Utilization

http://www.cisco.com/en/US/products/hw/switches/ps607/products_tech_note09186a0080094e78.shtml#baseline

HTH,

GP

[bbaillie]

The usual explanation I have run into for high CPU utilization on Cisco switches is its use of PVST+. This isn't a flaw just something that you need to consider when configuring these switches. The issue is the switch maintains a separate spanning tree instance for each VLAN on each configured trunk port. If you use the command "show spanningtree summary you will notice the total spanning tree instances on the switch. Maintaining soo many is CPU expensive. Use allowed lists on the trunk ports to ensure the trunks aren't required to support instances for all VLANS on the switch but just the required ones for the trunk port itself. Use the command "switchport trunk allowed vlan 1,22" where this port propagates VLAN 1 and 22.

Cheers,

Brian