VSM Performance

[hleschin]

Hello Cisco,

I have a question about VSM performance.

In our largest n1v environment (44 hosts,  432 vnic, 172vmnic) we see that the VSM (version 4.2(1)SV1(4a) ) has often a very high CPU load. The vcenter shows that the average VSM load is always ~75% , with many 100% peaks.

The VSM view shows a little different values, here is the average load sometime also 100%, but the general load seems to be lower than the values we can see in vcenter.

We are a little worried, if the high cpu load can have any impact for the environment (keepalive handling ...). We plan to add 14 more servers in the next months. The question is, if it's possible to give the VSM more CPU resources ? It seems for me, that multiple cores are currently not supported. Could you confirm that ?

If so, is it planned to change this in future releases?

Regards

Hendrik

[Robert Burns]

Hendrik,

As the 1000v utilizes distributed processing, there should be not be any issues running with that scale.  The 1000v supports up to the following scaling limits at your version (1.4a).

64 VEM per VSM

16K MACs per VEM

216 vEths per VEM

32 Eths (vmnics) per VEM

You're no where near these limits currently.

Are you using any other features by chance?  NetFlow, SNMP, QoS, ACLs etc?

All the actually switching and processing is offloaded to each VEM, so the VSM's processing should not impact your virtual networking.  All the VSM does is push configuration out to each VEM, and from there processing is handled locally.  For this reason there's no need for additional CPU or memory on the VSM VM.

I wouldn't be so concerned with what vCenter reports the VM utilization on.  If you want to monitor CPU/Memory usage, use the native monitoring from the VSM directly:

show system resources

Regards,

Robert

[Robert Burns]

Hendrik,

I did dig up a bug that affects your version.  It causes the VSM to report incorrect CPU history usage.

We're tracking this as CSCtr99730- show process cpu history command not giving correct data

This has been corrected in version 1.5.  This issue is cosmetic only and does not affect the performance/operation of the VSM.

Regards,

Robert

[hleschin]

Robert,

thanks for information. We had planed to start the upgrade to 1.5.1 this week. It's good to know that we will fix also this bug with the upgrade.

However, the special features we use are snmp polling and QoS . We can see, that especially the snmp polling causes mostly the high cpu load. Sometimes it happens during high CPU load, that the VSM can't answer fast enough for the snmp requests from our management systems. That causes unreachable alarms.

It's difficult for us to minimze the snmp requests, because we need the statistic information about load and errors for the vmnics + vnics. So I'm still hoping, that it will be possible in the future to increase the CPU resources for the VSMs.

Regards

Hendrik