Sup-7 cores backgroud processing

r.kukreja · ‎07-18-2011

Hi,

We recently installed the sup 7-E there are two cores on Ist core cpu utilization is 10% and on other one it is 15 % .can any one tell on what basis the cpu processing decision is made and why it is lesson Ist core and high on other core.

Regards,

Rajat

r.kukreja · ‎07-18-2011

Can any body share there view on above. its not lesson it is less on ist core and high on othercore as mentioned above.

Regards,

rajat

r.kukreja · ‎07-19-2011

can any body reply ?response will be highly appreciated

Regards,

Rajat

Jon Marshall · ‎07-19-2011

Rajat

It depends on how you have it setup and how the traffic passes through the switches. For example if you have HSRP and STP setup so that one of the switches is the active HSRP and STP root then more traffic will go to this switch than the other one.

So the fact that one switch is showing higher CPU than the other is not necessarily an issue. Like i say it's difficult to give any more information.

Jon

r.kukreja · ‎07-20-2011

Hi Jon,

Thanks for your exclusive response. actually StP and HSRP kind of traffic is not flowing through this swtich it is in transparent mode.

my qusetion is if there are 10 packets why seven packet go to ist core and 3 packet to 2 core. why dont packet equally get distribute.

Regards,

Rajat

Jon Marshall · ‎07-20-2011

It all depends on how your topology is setup. Without knowing how the switches are interconnected it's impossible to say. It could simply be that you have more devices communicating via one switch than the other.

Like i say, without understanding the topology it's impossible to say.

Jon

r.kukreja · ‎07-21-2011

There is not an issue of topolgy. there are SAP servers and database server and web servers are connected just

and my question is not topology specific. its a general question whose answer i am eager to know. I just want to know

why 2 cores does not have same utilization at same period.

Jon Marshall · ‎07-21-2011

But it is to do with our topology. For example lets say you have multiple servers dual honed to both switches. If you are running active/passive then the switch with the active connection will be receiving more packets than the switch with the passive connection.

It is all to do with how much traffic each switch is processing and unless you can confirm that each switch has exactly the same number of active connections from servers/uplinks etc. then it's not surprising that one switch has a higher utilisation than the other.

Jon

edvgispteam · ‎07-25-2011

hi,

last week i replaced our campus switches (2 4k5 sup6) with sup7 and see the same "issue" as r.kukreja. Of course the switch which is the root bridge and hsrp primary got more cpu utilization than the other(that was already like that with the sup6) but we are talking about 50-60% more utilization( and approximatley 20% more than with the sup6). when i do an "show proc cpu sort" i see that the load isn´t equally load-shared at all, my question is what happens if one of the cores hit´s 100% or some watermark at 90%? Will then start some sort of intelligent load-share process? Can i influence that behaviour with some hidden command other than making 50% of my hsrp&stp instances primary on the other switch?

Core 0: CPU utilization for five seconds: 25%; one minute: 26%; five minutes: 25%

Core 1: CPU utilization for five seconds: 74%; one minute: 65%; five minutes: 64%

PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process

10151 1533830 37678302 1 43.27734 40.64355 39.88671 0 iosd

5474 1111163 5329219 574 0.353516 0.311523 0.315430 0 ffm

10965 167388 1058260 3949 0.040039 0.040039 0.040039 0 cpumemd

10109 16624 872705 4284 0.017578 0.002930 0.001953 0 licensed

8326 12321 859260 4342 0.011719 0.002930 0.002930 0 cli_agent

5450 12519 953909 4119 0.002930 0.002930 0.002930 0 hwcontrol

5460 10131 846035 1821 0.002930 0.002930 0.002930 0 profiled

5480 17008 962825 4282 0.002930 0.003906 0.003906 0 plogd

5547 14815 954105 2023 0.002930 0.002930 0.002930 0 pdsd

8347 13276 1058604 369 0.002930 0.002930 0.002930 0 ngdumper_provid

10971 11002 882531 2733 0.002930 0.002930 0.002930 0 licenseagentd

8 7818 324919 10843 0.001953 0.001953 0.001953 0 events/1

5438 10460 859752 2175 0.001953 0.001953 0.001953 0 ha_mgr

5442 10087 845189 1771 0.001953 0.001953 0.001953 0 oscore_p

5454 10328 845036 2057 0.001953 0.001953 0.001953 0 ns_oir_proxy

5456 10353 848217 2079 0.001953 0.001953 0.001953 0 sysmgr

5466 10301 846231 2022 0.001953 0.001953 0.001953 0 os_info_p

5470 13831 937382 1009 0.001953 0.001953 0.001953 0 eicored

5602 10114 849553 1794 0.001953 0.001953 0.001953 0 iifd

7972 10565 858390 2301 0.001953 0.001953 0.001953 0 dtmgr

8239 10733 856520 2503 0.001953 0.001953 0.001953 0 installer

8257 10402 850223 2131 0.001953 0.001953 0.001953 0 snmp_subagent

9731 10202 845370 1907 0.001953 0.001953 0.001953 0 netd

10533 9986 846620 1649 0.001953 0.001953 0.001953 0 liin_tap

7 5971 323381 5184 0.000977 0.000977 0.000977 0 events/0

thanks

Gernot

Jon Marshall · ‎07-25-2011

Gernot

If the switch hits 90 - 100% it will not intelligently load-balance although you may find that it cannot respond to HSRP queries and so the standby becomes active.

If it is a concern then you should look to load balance your vlans as you suggest with odd vlans on sw1 and even vlans on sw2. That would at least spread the load from the clients. But you also need to be aware of return traffic if the traffic goes beyond the 4500 switches ie. if you had a WAN router one connected to sw1 that would be an issue as well.

Jon

edvgispteam · ‎07-25-2011

Hi Jon,

thanks for your quick reply. alright, so i think the answer is that i have no choice in influencing that behaviour(except for changing my topology design). One last question, what´s the point of having a multicore environment when one core is doing almost all the work and the other one is "sleeping", do you know what´s behind those iosd processes that consume most cpu time?

thanks!

Gernot

Jon Marshall · ‎07-25-2011

Gernot

Having a core with 2 switches is primarily for redundancy so that if one switch fails there is no interruption in service within your network. That is the primary reason.

However, switches, especially the modular ones, are expensive so it makes sense to try and use both if you can. That is why a common approach is to spread the STP root and HSRP active between the switches that do the inter-vlan routing. And if you hve WAN connectivity it is also a good idea to connect any WAN routers to both switches, that way each router has 2 active paths for return traffic and will spread the traffic load.

Note that obviously within a network you never get a 50/50 spread of traffic because some connections send a lot more traffic than others but it is still worth load-balancing vlans on our core/distro switches.

Jon

edvgispteam · ‎07-25-2011

Hi,

i think we are talking about different things when we use the word "core". when i wrote core i meant a single core from an dual core cpu as the one in the sup7. take a look at the output(that´s from one physical switch with a single supervisor engine in the chassis):

Core 0: CPU utilization for five seconds: 25%; one minute: 26%; five minutes: 25%

Core 1: CPU utilization for five seconds: 74%; one minute: 65%; five minutes: 64%

It seems that when you are talking about core you mean an switch as part of the core switching layer, right?

thanks :-)

Gernot

Jon Marshall · ‎07-25-2011

Gernot

Good point

I suspect, but do not know, that some processes/protocols might well be assigned to a specific core ie. they always use the same CPU and that is why you are seeing a higher utilisation on one than the other but i can't say for sure.

Apologies for my misreading of the question.

Jon

edvgispteam · ‎07-25-2011

no problem :-)

i think i will open an TAC case, when i got more informatin on that i will update this thread!

thanks anyway

Gernot