Different amounts of RAM between ESX hosts

[Jim_Stiveson]

I have seen some discussion on this. I am very much of the mind that you are better off with identical hosts. However we are going to be upgrading our hosts to new hardware soon and with the bundle we are being sold, two of the hosts will have 192GB of RAM while the other four will have 96GB. I would really appreciate hearing any and all thoughts on this. Does anyone think this is a show stopper? A non-trvial issue? Or, no big deal.

Thank you ...

[DSTAVERT]

Is there a reason for the large difference? Will they form two different clusters??? A little background.

[Jim_Stiveson]

My apologies. I should have provided more information. It will be in a single cluster of six hosts. We are purchasing a UCS bundle from Cisco and this was how the bundle came up for the best pricing. Don't ask me why ... LOL:smileyconfused:

[DSTAVERT]

If all the hosts had 96GB RAM would it be sufficient for your needs? If so I wouldn't be concerned. Can the extra RAM in the 2 hosts be spread between them all to balance the amount?  ????

[Jim_Stiveson]

It would be sufficient. We have been told that we can not spread the RAM around due to the board architecture so we would be stuck either as is or removing the extra RAM and leaving it on the shelf.

[DSTAVERT]

You could run into issues if you really started to make use of the extra RAM capacity in the two hosts. You could have problems with HA. I would probably remove the extra so that you don't have the opportunity.

[Jim_Stiveson]

I greatly appreciate you taking the time to respond. I am thinking the same way myself.

[DSTAVERT]

Very welcome.

[AureusStone]

I disagree.  I feel it would be better to use the additional memory, but keep in mind the issues with mismatched RAM.

DRS will try to balance memory usage based on percentage used, not free MB.  So if you are using 95% RAM the 96GB blades would have 4.8GB free, while the larger blades would have 9.6GB.

If there is an HA event on the 192GB blades you would need to ensure you have sufficient capacity on the other blades.

Later down the track if you expand and require more memory, you won't have to upgrade as many blades.

[Jim_Stiveson]

I can see your point also. However it is just the HA event you mention that I wish to avoid. Other than that, it doesn't sound like it is too big of an issue.

[bulletprooffool]

I agree with this

If you are settinhg your environment up so that you can tolerate loss of a server when all servers are 92GB and you do not actually provision more than VMs than you would have, had you had only 92GB per ESX host, you will not have HAS issues - and you'll have the bonus of extra physical memory for your running environment.

Secondly, when I design clusters, wherevere possible, I design them as N+2 (ie . .can tolerate 2 failed hosts) - If you are able to do this, you will be able to easily tolerate loss of one of your hosts with more RAM.

Lastly, I would add the extra physical memory, so thta in future . .when budget allows I can simply upgrade memory on the remaining hosts, rather than having to upgrade RAM on all hosts.

I regularly run mismatched RAM allocations in clusters, when running through upgrades of hosts.