3750 usage question

Unanswered Question

I want to ask a question regarding using a 3750 stack as server connection stack instead of plugging directly into the 6509 core. We have 2 6509's with sup32's and 5 1gig copper blades. Instead of using he copper blades it was decided to put in a stack of 3750's for the server connections and connect them to the 6509 via fibre.


I have noticed some spikes lately to 90% cpu utilization during the times when Ciscoworks and Orion are accessing the switches via snmp.


thanks


  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 4 (4 ratings)
Loading.
xcz504d1114 Sun, 12/30/2007 - 09:10
User Badges:
  • Bronze, 100 points or more

How many servers do you have connected to the 3750's? How many different networks? Are the 3750's operating as L3 switches?


And depending on the model of 3750 you could have a throughput of 6.5 mpps or 38 mpps... Depending on the number of servers, services running and the model of 3750 you have installed will affect your network traffic.


Typically 3750's are used for edge devices not data centers or core switches.


Here is a list of model types and their capabilities.

http://www.cisco.com/en/US/products/hw/switches/ps5023/prod_models_comparison.html


Something smaller than a 6509 but able to support a data center is the 4948.


Good luck.

johgill Sat, 01/05/2008 - 12:56
User Badges:
  • Bronze, 100 points or more

Honestly, none of this matters - the heart of the matter is CPU, and on a switch that should not be used much at all. A 3750 stack essentially gets you to a classic (bus) 6500 and the different models of 3750 give you different speeds, but talking about PPS here is marketing semantics. Different models have different numbers of ports and the main constraint is your stack speed - 16Gb/s.


We should not be using CPU at all for most traffic going through this switch, so the throughput discussion is a distraction here.


Check "sh proc cpu sorted" when this happens to determine where we should go from here. Is it an actual SNMP process or something else? Look for interrupt vs. process utilization as well. When it says something like 10%/6% that would be 6% of the overall is used by interrupt and the remaining 4% is explained by the processes following.


If it's mostly interrupt, that would be driven by traffic usually, and the first step there is to carefully watch "sh controllers cpu-interface" and we can go from there.


If it's mostly a process, what process is doing this and how often?

chschroe Sat, 01/05/2008 - 13:03
User Badges:

John is quite right, the CPU doesn't do the packet forwarding.


That being said, a Stack of 3750s has a lot of blinking lights and things to manage, and it uses just one of the CPUs in the stack to do that, plus getting the control plane traffic to and from that CPU, which could be far away.


Expecting SNMP polls to consume a lot of CPU on a stack of 3750s is not out of the ordinary. There's not much you can do about it on the 3750 platform unfortunately.


NS

Actions

This Discussion