How many VMs to run on ESXi hosts?

[Mike-S]

Hi,

I am after some advice on VM to host ratios. I am currently running around 30 : 1 (30 VMs to a single host). My data centre has limited space and with servers now capable of running 1TB of memory I could potentially place hundreds of VMs on a single host.

However, if I go down this path and have a host fail I would potentially have an outage of 100 servers while waiting for HA to restart them on another host. This would be a sufficent impact to the business.

I am after some advice from other people on comfort levels with VM to host ratios? This is really a decision around risk and not a technical decision. I want to understand what criteria you have used to put a limit on VM to host ratios?

Do you know of any published articles that talk about this topic? I need to have some supporting evidence to put a cap of the ratio but an interested to understand what others have done.

thanks

[idle-jam]

it depends how powerful your VM utilization (and also then number of vCPU required) and also how largely configured your hosts are (2CPU,4CPU or 8CPU with maybe 12 cores). If you have the current workload running, you can run vmware capacity planner to know and how to size it up as well as reading the vmware HA slots and how it affect the failover. http://kb.vmware.com/kb/1010594

[AndreTheGiant]

There are several cluster feature, like VMware HA and VMware FT, to increase HA.

But 100 VMs on a single hosts require also network and storage bandwidth and a lot of RAM.

IMHO... you have space constraints, consider to use blade solution or rack with 1U...

Or some vendor has now some specific solution (for example Dell has PowerEdge-C series) to have high density.

Andre

[Mike-S]

I am not sure if I made myself clear in my original post. I am not interested in the technical aspects as I know all about them.

I am interested in the risk factor with hosting many VMs on a single server. In the old days if a single server fails in your data centre you have a single system outage. If a ESXi host fails with 100 VMs I have a 100 system outage until HA restarts the VMs on another host.

I want to know what ratios of VMs to hosts companies are using and what influenced their decision? I also want to know any good articles that explain the business risk side of things.

[sergeadam]

The answer is always 'it depends'.

First, what are the VMs doing. The one thing I've noticed is that since most software now insists on being managed through a browser it is easier to create a number of small single purpose VMs. Those in a lot of cases are not critical. Windows Update, anti viru, network monitors and the like can usually sufer being offline for some time, therefore you can host a lot of those on a single server.

Then look at your hardware. Redundant power supplies, redundant NICs make for machines that can withstand some level of failures.

Bear in mind that hardware does not fail nearly as often as it used to.

Factor in the cost of having workers idle while HA kicks in versus the cost of additional hardware and licenses. 

[niv201110141]

@Mike I m sending you some configuration maximums which is recommended from vmware.Hope thats what you are looking for

For ESX/ESXi 3.5: http://www.vmware.com/pdf/vi3_35/esx_3/r35/vi3_35_25_config_max.pdf

For ESX/ESXi 4.0: http://www.vmware.com/pdf/vsphere4/r40/vsp_40_config_max.pdf

For ESX/ESXi 4.1: http://www.vmware.com/pdf/vsphere4/r41/vsp_41_config_max.pdf

[petedr]

Here are a few articles on the topic.

http://www.virtualizationpractice.com/blog/?p=7744

http://www.itbusinessedge.com/cm/blogs/cole/how-many-vms-is-too-many/?cs=34623

http://searchservervirtualization.techtarget.com/tip/Sizing-server-hardware-for-virtual-machines

[bulletprooffool]

Mike - I run:

Production at about 20:1

Development / Staging 30:1

Workstations 50:1

In addition, my lusters tend to be relatively large (at least 6 hosts) and have enough capacity for any 2 hosts in a cluster to fail at any point - and all VMs still to recover - without the hosts running at over 75% CPU / Memory

The more hosts you have in a cluster, the smaller the impact of a lost host - so the N+2(ability to lose 2 hosts) is more important than the number of VMs per host.