Imbalanced DRS cluster no recomendations.

[nscenter]

Good day everyone.

We a have 2 hosts DRS clusters, one host failed and HA did restart most of the VMs on the working host.  After restarting the failed host DRS shows no recomendations and besides the cluster is clearly imbalanced it looks like DRS does not agree with that.  So almost all the workload is running on one host only.   I wonder why DRS did not work this time?

Attached two snapshots, Thanks in advance for your help.

[JPM300]

Hey nscenter,

Usally when DRS is unbalanced and its not due to host constraints it is usally due to Reservations, CPU Affinity, or DRS host / VM rules forcing things to stay on a host, or making sure VM's stay together.

Check to make sure none of these things are in effect.

[vfk]

also, just to add JPM300 point, DRS also into account the cost of the migration vs the benefit of the migrating the vm to another host, if the cost of migrating the vm is higher than the perceived benefit VM are not moved, and sometimes you will observe imbalance, but it should correct itself overtime.

[nscenter]

thanks for fast and helpful reply,

in this case there is no affinity rules for the vms tha were moved to the working host.  I guess DRS believes that the cluster its ok that way.

I did manually the load balance, but still I'll remaing with the doudbt of what consider DRS to, this case make recomendations, for imbalanced clusters.

Thanks both for your help.

[vfk]

You can use DRS rules to separate VMs of the same functions so  that they are never of the same host, other than set it forget it.  Also looking at you screen your screen shots again I have noticed you have DRS set partially automated, it will not move VM in this settings, it will only make recommendations.  You have to set it to fully automated for DRS to automatically balance your cluster.  Hence pending recommendation.

Partially Automated : Initial placement is performed automatically. Migration recommendations are displayed, but do not run.

VMware vSphere 5.1

[Cyberfed27]

Our DRS does the same (sort of). We will have some hosts with 80% memory usage on while others sit at 40-50%. I would think DRS would balance that out but it doesnt. It has to get real bad before DRS kicks off an automatic migration.

[nscenter]

right, I agree Cyberfed27.   In this case we rather to set it to partially automated so we can apply the recomendations, but as you point it seems DRS will only work when one host is almost using all its resources without taking in account other (kinda idle) hosts with plenty of resources avaliable.

[JPM300]

This is probably due to the fact that even with the 1 host running everything there is little to no conention so DRS see's no performance gain in moving the VM so it doesn't reccoemend it.  If you put more load on that one host you would probably see the reccomendation happen.  You can also run the DRS check manually by clicking on the Cluster > DRS tab > Run DRS

Hope this has helped.

[vfk]

This depends on what you are comfortable with, in my own environment I have DRS fully automated, and DRS rules to separate critical VMs.  If you have to apply every recommendation, you have to constantly watch your cluster for imbalance and this could hurt your performance as one host could be overloaded while the other hosts sits idle...waiting for DRS recommendation to be applied. 

DRS and vMotion are very stable technologies, try fully automated and get comfortable with the settings...

[weinstein5]

Also remember DRS only kicks in if there is resource contention - so if the host where the VMs restarted has sufficient resources to support the CPU and Memory demand the VMs will stay there until that host is not able to provide the resources -

[vThinkBeyondVM]

Agree with most the comments by all.

2 more important observations I would like to highlight is " Migration threshold" :

As per the screen shot: Migration threshold set on your cluster is : Apply priority 1, 2, 3 recommendations. It is mean that DRS will only recommend migrations those will promise at least good improvement to the cluster's load balance. If you go ahead and make migration threshold more "aggressive", DRS will recommend the VM migrations to other host.

Also, as you have HA also configured on your cluster. As per the admission control, HA has to keep some cluster resources free in case on host failure, in that case also we can not expect DRS recommendations.

[jerzm]

The problem is that vCenter will let host alarms trigger for memory utilization being too high on a host, but it won't also then migrate via DRS.  This isn't at all what it should be, or give us the option to disregard contention and keep host resources actually balanced, which I think is what most people want anyway.