Adding memory to host

[tractng]

Hello,

I need some feedback here.  We have 5 hosts (each has about 128 GB memory) with a total of 96 VMs.  We need to bring each host down to add memory (can be done different days).  I generated a report from Veeam that tells me which host is the ideal candidate to bring down plus other factor such as the VMs could be shutdown during the process.

Host 1 has 19 VMs and is a candidate to take down first.  9 VMs can be shutdown during the whole process.

Here is my thought process:

1) Shutdown the 9 VMs on host 1

1A) Set HA to not shuffle the machines around since I am doing this manually.

2) VMotion 10 VMs to other hosts

     -  take down other VMs on other hosts to lessen resources (on other hosts, I also have the ability to take down a few VMs)

QUESTION: As the VMs are shutdown, the "Memory Usage" will reduce correct for each host?  But I still have to make sure the VMs total memory allocation is lower than the limit of 128 GB event hough some machines are powered down?

3) Put in maintenance mode and shutdown host

4) Add memory

5) Start host and turn on any down machines, VMotion, etc

I can have one host with the new added memory run (greater than other hosts) for days correct?  The hosts don't need to have equal memory?

[npadmani]

QUESTION: As the VMs are shutdown, the "Memory Usage" will reduce correct for each host?  But I still have to make sure the VMs total memory allocation is lower than the limit of 128 GB event hough some machines are powered down?

That's right, with every VM you shutdown it will reduce resource utilisation from available resources. you have mentioned HA in your discussion, but there's no mention of DRS, is it also enabled on your cluster? if so, best way to go about is, try to enter host in maintenance mode (disable HA Adminission control policy), DRS will help you to find best possible hosts among remaining 4 in your environment and do live migration of VMs from the host which you are entering in maintenance mode.

if you see that available resource are not enough to support all the VMs to migrate then you can make a decision to shutdown some VMs.

Total memory allocation can be overcommitted, but might give you negative impact on performance if services are going in higher utilisation, that's why earlier I mentioned that let the DRS do this for you.

I can have one host with the new added memory run (greater than other hosts) for days correct?  The hosts don't need to have equal memory?

yes, your cluster will still be operational. won't be a good idea to run your cluster like that for long but in your case since you are upgrading RAM in your hosts, should be ok for a day or two.

[tractng]

npadmaninpadmaninpadmaninpadmaninpadmaninpadmani,

Thank you for taking the time to provide me with the details.  Yes, we have DRS turn on.

So in practice, If we don't shutdown any VM machines and put a host into maintenance mode, it will start to move all powered on VMs to other hosts?  But if I shutdown the VMs, DRS will not move the off machine to other hosts? 

Can you explain why HA is disabled. I read here on the monitoring part.  I guess HA will cause a lot of issues with resources not met, alerts.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=203354...

I am in the middle of rebuilding my test VMware environment so I am asking here.

Thanks again.

TT

[npadmani]

If we don't shutdown any VM machines and put a host into maintenance mode, it will start to move all powered on VMs to other hosts?  But if I shutdown the VMs, DRS will not move the off machine to other hosts?

if your DRS is set as Fully Auto. and your environment is meeting all the requirements of vMotion compatibilities, then powered on VMs starts to move away from the host which is entering in maintenance mode. But don't worry about powered off VMs, when you try to enter host in maintenance mode, it will prompt you with question that Would you also like to move powered off VMs and Templates away from the host? give your consent to it and powered off VMs will also move.

Can you explain why HA is disabled. I read here on the monitoring part.  I guess HA will cause a lot of issues with resources not met, alerts.

I guess you are asking that why you have to disable HA Host Monitoring while this maintenance activity? if so, the KB which you are referring is about planned maintenance happening with your Management network (which is HA heartbeat network), if you don't disable HA Host Monitoring, it will cause Isolation Response to be triggered in some hosts.

While in your case, you are doing host maintenance activity, in this case, no need to disable HA or Host Monitoring, let them continue with normal operations. you might have to consider disabling Admission Control policy of HA so it doesn't stop you placing host in maintenance mode.

[tractng]

Right now the DRS is set to "partially".  I assumed if we set to fully, once we put a host in maintenance mode, it will vMotion the machines right away (take a walk and coffee break).

Thanks,
TT

[DeepakNegi420]

My response in green-

Since you have mentioned HA so I'm assuming you have adequate memory to perform this activity without downtime. Do you have HA errors to satisfy HA (As it calculates the slots) in your cluster? I'd suggest you to keep DRS fully automated and put one of the hosts into maintenance mode and it will automatically migrate the VMs off and also you need to monitor the utilization of your hosts, ideally DRS should be able to take care of it and ESXi will not move to maintenance mode until all Vms are off from it.

Hello,

I need some feedback here.  We have 5 hosts (each has about 128 GB memory) with a total of 96 VMs.  We need to bring each host down to add memory (can be done different days).  I generated a report from Veeam that tells me which host is the ideal candidate to bring down plus other factor such as the VMs could be shutdown during the process.

Host 1 has 19 VMs and is a candidate to take down first.  9 VMs can be shutdown during the whole process.

Here is my thought process:

1) Shutdown the 9 VMs on host 1

You may not require to do this.

1A) Set HA to not shuffle the machines around since I am doing this manually.

HA has nothing to do with the vMotion.

2) VMotion 10 VMs to other hosts

     -  take down other VMs on other hosts to lessen resources (on other hosts, I also have the ability to take down a few VMs)

DRS should be able to manage the workload across all ESXi hosts. 

QUESTION: As the VMs are shutdown, the "Memory Usage" will reduce correct for each host?  But I still have to make sure the VMs total memory allocation is lower than the limit of 128 GB event hough some machines are powered down?

Memory usage will reduce when the VMs are off the host in form of migration or shutdown state.

Use DRS and following approach is good then,

3) Put in maintenance mode and shutdown host

4) Add memory

5) Start host and turn on any down machines, VMotion, etc

I can have one host with the new added memory run (greater than other hosts) for days correct?  The hosts don't need to have equal memory?

You can have one hosts with more memory in a cluster however it's recommended to keep the same amount of memory in all of the hosts. For a limited time it should not really matter if one hosts has more memory.

[tractng]

Thanks guys.  Been away since we have other projects. .

Currently we have DRS to "partially".  When we set DRS to "fully", how soon will it start to shuffle things around (100 VMs on 5 hosts environment)?  This will give us an idea when to make the setting change and shutdown any VMs prior. We plan to do this mid March.

Here is an updated outline:

1) Shutdown a few VMs on host 1

      -

2) Disable control policy in HA

3) Set DRS to "fully" a few hours before host 1 is shutdown

4) Put host 1 in maintenance mode and shutdown

5) Add memory

6) Start host 1 and turn on any powered off machines

** Second day **

Put host 2 into maintenance mode and let DRS do its thing, shutdown host and add memory (same with the rest of hosts 3-5)