Resource Pools Caveats
http://www.yellow-bricks.com/2010/02/22/the-resource-pool-priority-pie-paradox/
After reading the following blog posts from
www.yellow-bricks.com and doing my presentation at the VMUG (2009) I wanted to show you some screenshots to emphasize the problem a bit more using real-life screenshots.
http://www.yellow-bricks.com/2009/11/13/resource-pools-and-shares/
By the way, use these links to read more about this topic:
AN EXAMPLE
CPU
Let's take a look at this example.
We have a datacenter, 4 ESX servers and
2 resource pools (Production & Test)
Production is very important to us, so share
setting is set to: HIGH.
Test is less important to us, so the share
setting is set to: LOW.
Where LOW:NORMAL:HIGH = 1:2:4,
Production gets 80% of shares,
Test gets 20%.
Production
First let's take a look at the 'Production' resource pool. Take in account we have 4 ESX servers, each having 4x3Ghz (so that's 48Ghz in total). With 80% of those resources for 'Production', we will not see any problems, so 'Worst Case Allocation' will get 3Ghz for each of my VMs.
Test
Next take a look at the 'Test' resource pool. We also have a share setting on these VMs.

Because the 20% of shares set on 'Temp', there will be some resource contention. This is shown in the 'Worst Case Allocation'.
Now keep in mind this 'Worst Case Scenario' takes in account ALL VMs and ALL resource pools in THIS cluster. So CPU on EVERYTHING goes to 100% (in this case)
!
!
Let's take a look at the following: A lot of our tests passed, and now moved over to 'Production'. That means there is only 1 VM left in 'Test'.
So while the 'Test' resouce pool is still at 20% shares, and even while the VM itself has the share setting to 'Low', the 'Worst Case Allocation' is 3Ghz!!!
But what happened with our 'Production' VMs? Let's take a look and take a close look at the 'Worst Case Allocation'!
So a 'Low' VM in 'Test' can get 3Ghz, while a 'Normal' VM in 'Production' will only get 1,6Ghz in worst case... Confusing huh?
?
?
No it is not! It's logical - let's take a close look at what just happened...
This is 80% / 20% share allocation
The same here, but now the individual VMs per slice. Quite logical.
And here the change where 1 Test VM has the full 'Test' Resource Pool slice.
So, make sure you use the Resource Pools correctly and think about what the end result would be.

Also, do not forget to look at the 'Worst Case Allocation' on the Resource Pool overview Tab.
By the way, to see the pictures better:
click "More" and then "Fullscreen"