When Is It Time To Get A New Host

[fitzie22]

I have a drs/ha cluster with 2 hosts and the "current failover capacity" is still at 1. Now the problem I have is I need to be able to give notice on when to purchase a new server for the cluster. Is there a way to know when it will change from 1 to 0 and failover will not work?

[kjb007]

I'm not sure how aggressively you run your hosts. I myself will run weekly reports as to the utilization of the resources on the hosts. When you've exceeded in the actual utilization of your hosts vs the # of your failover capacity, it would be a good idea to get another host.

This is, of course, a simplified answer to your question. Overcommittment, future expected growth, current capacity, all come in to play when capacity planning for your environment.

-KjB

[fitzie22]

I myself will run weekly reports as to the utilization of the resources on the hosts "how do you run your reports? do you use a 3rd party utility?"

When you've exceeded in the actual utilization of your hosts vs the # of your failover capacity, it would be a good idea to get another host " I am really not sure what you mean with this. can you elaborate a little more?

[kjb007]

I don't currently use a third part. I run utilization reports (cpu/memory) on the cluster, at least once a week. The report will show your utilization, and you can use that over time to predict when the utilization, growing over your trended weekly reports, will ultimately exceed how much resource you have available. Between the two hosts, if your utilization exceeds 50% on each server, then you will have performance problems in an HA situation, and your machines will be swapping, and/or ballooning to maintain performance levels. Once I reached that utilization mark, it would be time for a new server. In a 3 host scenario, the number would be above 66%, and so on.

Remember HA calculations are based on slot size, and not actual vm resource utilization. You can see this thread for description on HA slot size calculation: http://communities.vmware.com/thread/154865

-KjB

[Anders_Gregerse]

It's always time to get a new host

The failover capacity is calculated on the worst case senario. It takes the biggest vm you have, slices your hosts up in "biggest vm" slices and count how many slots you have left. That doesn't mean that you have run out of failover capacity

I usually just do the beancounting to see if the used capacity (memory and cpu) is more than the free resources. If you got a large environment there are tools that can help you.

[fitzie22]

Perfect, thanks I will check that out. I know I definatelly need to get a new server in that cluster.

[kjb007 · Accepted solution]

No problem.

-KjB

Don't forget to leave points for helpful/correct posts.

[fitzie22]

when you say bigest how does it do that. is it teh vm with teh most memory assigned to it?

[kjb007]

Check the link I posted earlier about HA slot size. By biggest, it is biggest slot size. Slot size is calculated using memory/cpu reservations, and not allocations.

-KjB

[y_wisdom]

A good rule of thumb is to have enough spare capacity in your cluster so that you can shut down at least ONE host for maintenance. Unless you want to come weekends / nights / holidays to do all your maintenance.

Example

1. of host in cluster

2 Host = Minimum 50 % spare capacity on both hosts

3 Host = Minimum 33 % spare capacity on each hosts

4 Host = Minimum 25 % spare capacity on each hosts

5 Host = Minimum 20 % spare capacity on each hosts

...............

You got the picture !!!

[fitzie22]

Ok I have taken some perfomance stats from the prod cluster and as you can see from the memory usage pic that it is averaging 129 percent. now if you look at the failover capacity screenshot it is still cble to failiver. am I missing something?

[fitzie22]

memory image did not come through

[fitzie22]

one more time

[fitzie22]

OK I don't know why but sometimes I get an UNAUTHORIZED error uploading pics. REALLY A PAIN IN THE REAR

[kjb007]

If memory usage is 129%, look also at the memory balooning, and swapping. If you are doing a lot of that, then it will reiterate the point that it is time to add another host.

-KjB

[fitzie22]

I can;t tell if it is or not becaue when I select the cluster on teh left adn then the performancetab on the right the only data that is gives for memory is the %used. all teh other one's like swap adn balooning never pull data even if I say to just pull for today

[kjb007]

Select your host on the left.

Go to performance tab on the right.

Click 'Change Chart Options' link at the top

Click the memory link on the left, and then adjust your counters on the right hand side.

Then click ok, and your chart will be updated.

-KjB

[fitzie22]

Ok so I guess the only way to get balooning or swaping stat is to do realtime on teh host. anything else does not give me any stats. So it looks like one of my hosts in the cluseter is balooning at an average of 185250.2 kb. now the thing that confuses me is that the host is only using 19 out of 32 gig of memory. now the questiong is how to find the culpret that is doing it?

[kjb007]

That's interesting. Are you looking at a specific metric that is telling you that? Are you using reservations or any other memory management using resource pools or things like that?

-KjB

[fitzie22]

here is a screen shot. I can't upload pics so I hope you can read it. I am selecting teh host and memory and that is trying to pull data for a week. if I do real time it will pull all teh data. I am not using reservatinos but If you look at the second pic the previous vmadmin here setup reservation shares all wrong. I need to set them all to normal