scom agent vs scom management server

[nico.weytens]

A while back we've upgraded our collector servers from regular servers with a SCOM agent to full blown SCOM Management Servers. Performance wise that made a big difference: in stead of 5000 objects per collector we could target up to 12500 objects per collector. (2 vCPU/8GB RAM each)

Last week we've upgraded to v6 and I see the load is now measured in "workflows" instead of "objects". Each collector can handle up to 75000 workflows.
I'm wondering if we can still push that boundary with our Management Server collectors? Or doesn't that matter in v6?
We will be upgrading our SCOM environment to 2012 SP1 some time soon, and if collector performance with a regular agents is pretty much the same we wouldn't need to go through the trouble of setting up extra Management Servers.

[sergey.g]

Hi Nico,

First of all I would like to mention that in v6 we've improved performance a lot, a single agent should be able to handle way more "objects" now. We went to workflows instead of objects because all objects in virtual infrastructure are different - ESX host has one number of workflows, while VM has another number. In order to understand how big your enviroment is in terms of workflows you can use our deployment calculator

http://www.veeam.com/support/mp_deployment.html

Enter parameters of your infrastructure and check the HTML code of the page, there should be a hidden field, like this one:

Code: Select all

<p style="display:none" id="objects">591562</p>

This is a calculated (not actual, but very close to it) number of workflows for your infrastructure.

Now, each agent can handle up to 75000 workflows. If for your infrastructure management server could handle 2.5 times more, I beleive it's safe to assume that it should be able to handle more than 180K workflows, but I wouldn't promise that - this need to be double-checked for both of your environments: SCOM and Virtual.

With respect to comparing 2007R2 with 2012SP1 - in our labs, we haven't seen any significant differences in terms of performance.

So, as a summary: I would still recommend you to test v6 with a SCOM agent, we've made a lot of improvements for better performance and I think you may not need to use Management Servers on Collectors anymore. On the other hand Management Servers instead of agents could reduce number of necessary collectors significantly which is also a good thing. So it's up to you to decide and it depends on what you like more - more collector servers but with regular configuration or less collectors but with full management server software installed on them.

Let me know if you have any other questions.

Thanks.

[Alec King]

Hi Nico,

I have to correct and clarify some points above - we do not believe that any SCOM Health Service - whether Agent, OR Management Server - will achieve more than 75000 workflows.

We do recommend a Management Server for the largest environments, but not because of workflow count. A Management Server is more stable, better architected for large discoveries, and it performs less local disk I/O than an Agent (Management Server talks more-or-less direct to SCOM databases)

But for the number of workflows, 75K seems to be a real-world limit. It is a soft limit, but we found that exceeding this leads to very long discovery times, and if anything changes in the environment (such as even one VM vMotion), then the SCOM Health Service has to recalculate all workflows. When >75K workflows, this recalculation just eats CPU, memory and disk I/O, and effectively causes a monitoring outage which can be an hour or even more. So we don't recommend exceeding 75K workflows, even on Management Server.

I believe in your environment, probably you are already close to 75K workflows on each Collector? The performance counter is available in our MP, in view Veeam for VMware/ Veeam Collectors / Performance / Collector Load %. (this percentage is calculated as a % of the 75K recommendation). There are also views Total Veeam Workflows and Workflow Count in the same folder (these views show Veeam-only workflows, and the total workflows running in the Health Service).

I can say that MP v6, with it's redesigned topology, improved discoveries, and optimised rules and monitors (multiple events/metrics gathered by one workflow) is greatly more efficient than v5.7. If you use Management Servers, and MP v6, you should acheive the 75K limit; and this equates to approx 1000 VMs per Collector. This number varies depending on your exact topology - definitely recommend using the Deployment Calculator-
http://www.veeam.com/support/mp_deployment.html
which will give a good estimate on number of Collectors required for your specific environment.

Hope that helps - any questions let us know!

Cheers
Alec

[nico.weytens]

very in-depth and useful info Alec, thanks for that!

In our situation we have a smaller environment with 2 Agent/Collector servers, each have about 23000 workflows to handle. Problably we don't need to bother upgrading them to Management Servers. (although I currently have a support case open 'cause one of them is dropping data all the time)
For our big VMware environment we have 8 MS/Collectors, and soon we'll add 2 more. Currently each of these collectors are handling about 50.000 workflows so after reading your explanation we will probably benefit by keeping these as heavy-duty Management Servers when we upgrade to 2012 SP1.