policy section 3b: useable capacity thresholds. If there are 4 hosts, would a 25% threshold = N+1

[TedH256]

I'm trying to figure out what these thresholds mean. It seems to me that the % setting is going to mean something different depending on the number of available hosts (and this even will depend on an assumption that all hosts have equal resources).

I want to create a policy that will enable us to preserve a true N+1 capacity, such that when any one host goes down, the remaining hosts can fully accomodate additional workloads without loss of performance. Is setting these values to 25% going to accomplish what I hope?

What would the appropriate setting be if we had 8 hosts - 12 or 13%

[jddias]

Hi Ted,

  First, I note that you have the HA option selected - doing this will automatically use your cluster HA settings for the reserve of CPU and memory capacity (so in your case you are really reserving 50% not 25%).  Storage capacity is independent of the host capacity, so from the perspective of setting a failover capacity I don't think you need to worry about that.

  The reservation levels are, in my opinion, better for making sure that you have a buffer for unplanned growth than for HA.  I would use vSphere's clustering capability to ensure availability as your SLA requires and use the HA option in your policy to protect that (besides, there are other mechanisms in play with vSphere HA, such as admission control, that help protect that SLA).  Otherwise, use the buffers to give yourself some capacity for those last minute/unplanned VM requests.

[TedH256]

hmmm ... thank you for the reply, but now like many of the controls in vcops, I find myself wondering just what those words mean!

"First, I note that you have the HA option selected - doing this will automatically use your cluster HA settings for the reserve of CPU and memory capacity (so in your case you are really reserving 50% not 25%)."

We do not use HA admission control (just don't find it that useful) - but I don't understand what "so in your case you are really reserving 50% not 25%" could possibly mean? There are 4 hosts in the cluster. We try to maintain an N+1 capability, which would mean essentially each host has 25-33% excess capacity, so that if one goes down, all the workload can run. Where do you get 50%?

Also you said:

"The reservation levels are, in my opinion, better for making sure that you have a buffer for unplanned growth than for HA."

Huh? Yes, that is what we want - we would like vCOPs to let us know, when our actual useable capacity is close to not being N+1 - we don't expect vCOPs to have any role in actual HA situations, but we want it's capacity forecasting to understand that "N+1" is what we consider "at capacity" so that we need to scale up (or move workload to another cluster) ...

It seems hard to believe that what I am asking for is so difficult, as the notion of N+1 is fundamental to resource planning of a virtual infrastructure ....

[jddias]

If you want to consider HA, you only need to check the HA box in section 3b.  The other reservation levels are not needed if HA is your only concern, but to get that you will need to use Admission Control.

Otherwise, if you want, you can use the buffers but it won't give you the same level of protection that Admission Control does.

My 50% statement was based on your settings and my assumption that you WERE using Admission Control.  In other words, HA settings were configured to give you protection for a single host failure in a four host cluster (25%) and then you add a 25% capacity buffer on top of that = 50%.

[TedH256]

ahh --

so with "Use HA configuration" checked, if we were using Admission control, it would reserve 25% (CPU and Memory) on each host. Then the 25% that I put in the field, is an ADDITIONAL 25%.

So - the way things are (with no admission control configured) the "Use HA configuration" check-box does ... basically nothing, correct? So then - the 25% that I have set will essentially do what I was hoping it would do.

(I know, you don't think it offers the 'level of protection' that admission control does - but I don't need admission control - I just want to have a way of keeping track that the entire cluster is roughly N+1).

These "buffers" after all are not real - they are just calculation contrivances to let us know where we are with respect to capacity remaining, and using the "+1" as a floor is my only simple basic goal. Thanx for you help!

[jddias]

Right you are!  Sounds like you have a solid understanding.  Good luck!