Lower performance of a VM with FT?

[rickardnobel]

If running a VM with Fault Tolerance enabled, could I expect lower vCPU performance of this VM?

If the logging network is fast, say 1 Gbit/s, and the host has plenty of physical CPU resources, will the extra control of vCPU access and the logging still make the VM run slower?

[vmroyale]

Hello.

If running a VM with Fault Tolerance enabled, could I expect lower vCPU performance of this VM?

No, not at all.

If the logging network is fast, say 1 Gbit/s, and the host has plenty of physical CPU resources, will the extra control of vCPU access and the logging still make the VM run slower?

No. Check out the "[The Design and Evaluation of a Practical System for Fault-Tolerant Virtual Machines|http://www.vmware.com/files/pdf/partners/academic/fttech.pdf]" document. It details exactly how the lockstep technology works. This is the definitive guide to understanding how FT works.

Good Luck!

[chadwickking]

Hi,

Just wanted to throw in 2 cents and let you know in our testing/lab environment we haven't seen any performance problems with using FT. Most of the VM's that were already using 1vCPU and had FT enabled we didnt see ANY performance hit. I was actually quite surprised but its really a great product.

Some additional reading as well:

http://itknowledgeexchange.techtarget.com/virtualization-pro/masters-guide-to-vmware-fault-tolerance...

Also note that in vSpere 4.1 they added the ability for FT to be used with DRS. That was a functionality that it didnt have before. We haven't tested this but I would like to see some performance metrics.

Cheers,

Chad King

VCP-410 | Server+

"If you find this post helpful in anyway please award points as necessary"

Message was edited by: chadwickking

[AntonVZhbankov · Accepted solution]

>If running a VM with Fault Tolerance enabled, could I expect lower vCPU performance of this VM?

Yes, about 10% slower.

---

MCITP: SA, MCTS Hyper-V, VCP 3/4, VMware vExpert

http://blog.vadmin.ru

[AntonVZhbankov]

Brian, you wrote yourself: Overhead is dependent on the workload and can be as low as 5% or as much as 20%

---

MCITP: SA, MCTS Hyper-V, VCP 3/4, VMware vExpert

http://blog.vadmin.ru

[vmroyale]

I did, didn't I.

Not exactly sure how I managed to forget about the overhead, but Anton is absolutely correct. There is also a great blog entry over at the VMware's performance team VROOM! Blog that shows the performance using VMmark.

I guess the key here, as with most things, is to test your actual workloads in FT to see how they really perform.

[chadwickking]

It would be awesome to see an updated doc on the changes in FT in 4.1 like how ESX servers dont have to be the same build number and how it can run in DRS too. But just to comment on the performance question, my only guess as to why we didnt see a hit in our lab environment is probably due to the amount of use we may get in the lab. Or as you said "workload" just isn't as much in production. I would love to see the day when we decided MSCS was no longer an HA solution for us and that VMware could achieve the SLA we are looking for.

Cheers,

Chad King

VCP-410 | Server+

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[rickardnobel]

Thanks a lot for your answers. That some CPU performance could be lost is not that strange I guess.