Is my DRS balancing?

[caddo]

Hello,

for the sake of curiosity and to make sure i don't have a problem with my DRS i am posting a couple of pics regarding my DRS Cluster.

As you can see looks like it's removing all VMs from ESX4 for some reason. The config is default, if i make it more aggressive it removes even more VMs. I have no rules in place that could explain such behavior.

I'm running vSphere 4.1 Update 2.

Can anyone explain me why my DRS doesn't like to place VMs on that specific server? Why does it think it's balanced when i insted believe it's not?

Thank you.

[EdWilts]

Based on the current CPU and memory load, there's no reason to move any guests to esx4 since they're not constrained where they currently are.  If you put one of the other 3 hosts in maintenance mode, I would expect some guests to move to esx4 since the other 2 would now be memory constrained.

[caddo]

You have a point there, but how do you explain that if i change DRS settings to a more aggressive behavior the VMs get moved AWAY from ESX4?

It's a strange behavior since the other hosts are more loaded than ESX4, why would DRS move VMs from that host to the other ones?

I'm trying to justify such behavior or start troubleshooting it if it is a problem.

[caddo]

I tried something to further investigate: i put in maintenance mode one host and ths is the distribution.

Again DRS seems to priviledge hosts other than esx4.

[chriswahl]

Have you ensured that the VMs on the other hosts can be vMotioned to esx4 (datastores and networks exist)?

[caddo]

I know for sure because we did maintenace not so long ago and i can move all VMs on all hosts if i need.

[weinstein5]

Keep in my DRS does not 'balance' the load - DRS function is too insure approriate delivery of compute resources of memory and cpu to the VMs - so theoretically you could have a cluster of 3 hosts with 100 VMs and end up with 60+ VMs on a singles host if all 60+ vms were able to receive the necessary resources - 

[caddo]

Yes weinstein5, that's undoubtedly true but how constrained to the other hosts to be before moving VMs to ESX4? I mean, 94% sounds like high enough to start moving things around.

Other strange things happen, like if i manually move VMs to ESX4 then after few minutes DRS moves it back even if ESX4 is definitely undeutilized.

This is also why i think something could be wrong with DRS or that specific host.

[vGuy]

to ensure there is no historical information is cached and to clear it, you may try to:

restart the vCenter server service AND

disable/renable DRS (make sure you do not have any resource pools or DRS rules configured).

[Pogozelski]

Caddo,

DRS run after restart services vCenter ?

I'm having the same problem in ESX 4.1...

[caddo]

Your environment seems to be much more constrained than mine since i had half of my hosts doing almost nothing while you have all hosts over 75% of RAM usage but one which has more than 50% CPU usage.

Remember that balancing for DRS is not only about RAM but also CPU.

That said, i fixed my problem by putting in maintenance mode all hosts at turns, basically forcing DRS to shuffle all VMs on all hosts. After that the workloads were a lot more even throughout the whole cluster and the problem never showed up again. This allowed me to fix this with no downtime.

I can also see you have Admission Control disabled which means you could very well have turned on more VMs than you should have had; if this is the case DRS can't do much for you.

Hope it helps.