Comments on: vSphere cluster: max 4 ESX hosts per “location” because of HA limitations?

By: vSphere cluster: max 4 ESX hosts per “location” because of HA limitations? - Techblog

Fri, 26 Feb 2010 15:23:34 +0000

[...] http://virtualfuture.info/2010/02/vsphere-cluster-max-4-esx-hosts-per-location-because-of-ha-limitat... [...]

By: PiroNet

PiroNet — Fri, 05 Feb 2010 11:34:53 +0000

Duncan, will HA/DRS be rack/enclosure ‘aware’ to avoid situation where for instance cluster nodes get on same rack or worst on same enclosure? I think also DRS needs to be more ‘manageable’…
Thx,
Didier

By: Duncan

Duncan — Wed, 03 Feb 2010 22:28:41 +0000

As already stated I wrote about this a while back… feature request has been in place for a while and hopefully will be introduced in any of the upcoming versions. not much I am allowed to say about it unfortunately. (first link by the way when i google on “primary cluster design ha” .

Anyway, extra attention never hurts.

By: Matt

Matt — Wed, 03 Feb 2010 21:49:26 +0000

Does anyone know where additional command would be published? I would like to know what else it is capable of.

Matt

By: Arnim van Lieshout

Arnim van Lieshout — Wed, 03 Feb 2010 14:50:06 +0000

I ran into this when designing a cluster setup across 2 geographically dispersed datacenters.
Also see my post: http://www.van-lieshout.com/2009/11/geographically-dispersed-cluster-design/

I heart there are some VMware DRS and HA changes coming in this area.

By: Matthijs Haverink

Matthijs Haverink — Wed, 03 Feb 2010 11:54:08 +0000

Hi Ewan,

Thank you for your reaction, good point!

We searched with google and even on yellow-bricks itself and somehow we overlooked this article. It indeeds describes this behaviour but the fact that the article is from 1 year ago and still nothing has changed, gives me the feeling that it can use some attention anyway.

By: Chris Dearden

Chris Dearden — Wed, 03 Feb 2010 11:51:20 +0000

If you are limited by the number of chassis , then perhaps it would be a case for using more powerfull hosts ( quad sockets vs dual sockets for example )

By: Chris Dearden

Chris Dearden — Wed, 03 Feb 2010 11:50:09 +0000

I suppose it depends on how many blade chassis you are running ?

4 hosts per chassis per cluster isn’t an issue for me , as I have to seperate clusters by lifecycle anyhow – also have a good ammount of blades to play with.

if by location you move out to datacentre level, if I loose a whole datacenter , I’ll be failing over to BCP via SRM / other BCP techniques , so again its not a huge issue.

By: Ewan

Ewan — Wed, 03 Feb 2010 11:06:43 +0000

Duncan himself has covered this is another post than the ones you link to http://www.yellow-bricks.com/2009/02/09/blades-and-ha-cluster-design/

Basically it’s a bit of a pain but yes it’s a limitation – if you’re working across failure domains you probably want to keep your HA clusters down to 8 physical hosts.

By: bjorn bats

bjorn bats — Wed, 03 Feb 2010 10:36:56 +0000

very good post.
i a big environment we had the same problem.
what we did is separating the odd en even servers on the datacenters.

but then you have still limited primary nodes on every datacenter.