how many fault tolerance vm on single host

[akmic]

Dear all,

My questions is : if I need to deploy 30 virtual machines of low workload but important application so I need Fault Tolerance. Is it possible to deploy all 30 ft vistual machines on three ESX4.1 host and in case one ESX host is down. Some virtual machines will migrate to another ESX host in the DRS.

Does anyone has experience of many ft vm running on ESX host? 
It seems that vmware recommend no more than four fault tolerant virtual
machines (primaries or secondaries) on any single host.

Thanks

[notorious_bdg]

Configuration maximums document states that 4 FT VMs are supported per host.

http://www.vmware.com/pdf/vsphere4/r41/vsp_41_config_max.pdf

[akmic]

The vSphere Availability guide describe that the number of fault tolerant virtual machines that you can safely run on each host is based on the sizes and workloads of the ESX/ESXi host and virtual machines. And on page32 how to disable limit of the number of ft vm per host by the advanced option das.maxftvmsperhost in the DRS.

I think that the only 4 VM per host is too low consolidation ratio.
Probaby,I will ensure that vMotion and Fault Tolerance Logging NIC configured by 10Gb NIC. I think vmware recommendation of 4vm per host base on FT logging nic configured by gigabit ethernet, safty line of gigabit bandwidth, that is just my opinion.

I would like to know actual running of many ft vm on the single host.

Does someone work with many ft vm on one ESX host ?

thanks

[depping]

4 per host is what is supported according to our documentation. I would not recommend go beyond that if and when this workload is important to you and your customers.

Duncan (VCDX)

Available now on Amazon: vSphere 4.1 HA and DRS technical deepdive

[akmic]

I cannot read this book right now... However I cannot choice ft if this consolidation ratio.

thanks..

[idle-jam]

Hi,

i do not understand the last part.

[Gleed]

The "recommendation" of 4 FT protected VMs per host is based on having a 1Gbps logging network.  If you have a 10Gbps logging network you would likely have sucess running more than 4.  However, it's unlikely you would get the numbers you need (30VMs on 3 ESX hosts).

Keep in mind, with the FT record and replay nearly every I/O on the protected VMs will get replicated to the secondary VMs.  It doesn't take much to saturate the logging network.  Consider what happens if you were to patch multiple VMs concurrently, or just copy a bunch of large files onto the VMs - the sudden spike in traffic over the logging network would quickly saturate the link causing performance problems.

Also consider what would happen when you have a failover or put a host in maintenance mode, the number of VMs on the remaining hosts would increase significantly, which would result in not only more logging traffic on the NICs, but a lot more processing overhead as the record and replay will have to protect say 15 VMs vs. 10.

If you do pusue running more than the recommended 4 FT protected VMs per host I recommend you check with VMware support and confirm that you would have a supported configuration.

Regards,

-Kyle

[akmic]

I meant that only 4 vm per host is too low consolidation ratio. When i may deploy 30 vm for ft, i need 16 host and at least vSphere advanced 16 cpu license.
I think this solution is a little over budget. I may choose Stratus ftserver6300 with ESX standard..

thanks

[akmic]

I have understood your considerlation.
Many thanks!

And thank you everyone!

[aiea96701]

Thank you for this information.   Fortunately, when I consider FT, it is for my most critical VMs.   I do not forsee a need for more than 4 FT VMs.

btw...the discussion did prompt another question.   If I setup a single FT VM in a cluster, AND I require maintenance, how does that affect my FT VM.  I am sure it will fail to the secondary but once the host is back online, will it return.  I have a 3 host HA/DRS cluster.  So, I'm guessing that once a FT VM experiences a host failure , the secondary takes over primary and another secondary is created on the remaining host.   And, once the failed host returns to the cluster, the FT VMs do not move back.  

And, what if I only have a 2 host HA/DRS cluster (just thinking here).   The secondary VM would assume primary and I'd get an alert/error that no secondary exists on the new primary.   But, once the failed host returns.....does the "new" secondary get automatically created on the returned host?