VMware HA % based ACP

Oh man! Ad-blocking software has been detected! :'(

This website is run by the community, for the community... and it needs advertisements in order to keep running. Blocking our ads means your killing our stats!
Please disable your ad-block, or become a premium member to hide all advertisements and this notice.

Hi everyone, hoping someone can help me out a little with VMware HA % based ACP, here is a little run down of my lab setup.

Code:

2x ESXi Host
1x HA Cluster;
    31.98 GB RAM
    19 GHz CPU Resources
    Admin Control - Enabled
    Current CPU FO Capacity - 91 %
    Current Mem FO Capacity - 93 %
    Configured FO Capacity - 50 %
    HM - Enabled
    VMM - Enabled
    AM - Disabled

Current Usage;
ESXi1;
    CPU % 1
    MEM % 18
ESXi2;
    CPU % 2
    MEM % 28

Actual Guest MEM Usage - 22 % across the hosts
No Reservations on CPU or MEM for any VM

To me the current failover capacity doesn't seem to add up, in my head it would lead me to serious over-commitment of resources on the cluster.

 

Oh man! Ad-blocking software has been detected! :'(

This website is run by the community, for the community... and it needs advertisements in order to keep running. Blocking our ads means your killing our stats!
Please disable your ad-block, or become a premium member to hide all advertisements and this notice.

After reading a couple of articles I found out what the problem is, this feature could use some work

 

Does anyone use this type of admission control? The whole 0MB+Overhead(without reservations) seems like a silly design or am missing something.

 

Admission Control is about guaranteeing failover of VMs, and any reserved CPU/memory and memory overhead both need to be physically available in order to be able to guarantee to run/failover a VM.

I assume you've read the "HA Deepdive" over at http://www.yellow-bricks.com

Scott.

 

Sure did, but I can't get why it doesn't use actual usage on VMs instead of the 0MB+Overhead

e.g.

My vCenter VM has 2GB of provisioned memory and its using about 600MB at the moment with no reservation on CPU or memory and its overhead is 102MB

but when it's on percentage AC this VM would only require 0MB+102MB instead of 600MB+102MB. It seems really strange that I would have to manually calculate or set a common slot size when it should be able to track the memory itself.

 

The key word in my previous reply is "guarantee" - HA Admission Control will not guarantee that the regular or maximum/peak performance of your VMs will be unaffected by a host failure, but it will guarantee that your VMs will run at a minimum acceptable level of performance - hence the relationship between Admission Control and CPU/memory reservations.

Remember too that it's only the "host failures" Admission Control setting that uses slot sizes, the "percentage of resources" setting considers the reservations (plus memory overhead) of each individual VM rather than considering each VM as a "worse case" VM.

Scott.

 

Sorry slot was the wrong word to use, what I mean is setting das.vmMemoryMinMB. I think I'll need to order the new HA/DRS book lol

 

I'm afraid my reply this time has to be pretty much the same as last time - HA is all about guaranteeing availability, at a minimum acceptable level of performance, defined by your service levels for your applications which should influence the resources you reserve for your VMs.

Actual resource usage doesn't come in to any of HA's calculations, only reserved resources. If HA were to consider actual resource usage, the dynamic nature of actual usage would mean that the HA agents would have a lot of work to do on an ongoing basis to validate moments in time where there were sufficient failover resources available within the cluster, and where there weren't.

Remember also that HA "kicks in" independently of the availability of vCenter Server, yet it's vCenter Server that would hold all the actual usage data and be able to make the kind of validation I describe - in other words, HA works the way it does by design.

Who knows what the future holds for HA in subsequent vSphere releases, but that's where we are with it today.

Scott.