4 vCPU FT demoed with synchronous replication?
For a long time VMware Fault Tolerance was only supported on a single vCPU VM. Today we witnessed a change: During the BCO2874 session we got a first look on the prototype of the SMP enabled Fault Tolerance. Cool!
Introduction to FT
So what is Fault Tolerance (FT for short) all about? As most of you will know, FT enables you to have a VM running on one vSphere node, while a “shadow” of that VM runs on another host. If the host would die running the primary VM, the VM would resume seamlessly on the second host, now becoming the primary VM. FT would then proceed to immediately find a third host in the cluster to reenable FT and create a secondary (shadow) VM there.
FT has a number of drawbacks though. One of the biggest ones is that it only supports single vCPU VMs. A lot of people wondered why that would ever be useful; your primary and most important VM should be able to run multiple vCPUs. Up until now that has never been possible.
The Prototype demo
Beware, this is a prototype. There is no telling if or when this feature will be present in a future version of vSphere. Nevertheless, the prototype demo was cool!
The setup was as follows: Take a 4vCPU VM running an Oracle database. Enable FT on that host to a secondary host. Between these hosts build an FT-enabled VMkernel port with a 10GbE interface inbetween.
Next, load up the Oracle database with simulated load using Swingbench. As the VM was running, we observed 2-3Gbit bandwidth usage between both VMs.
Now that the setup was running stable, the host running the primary VM was hard rebooted. The primary VM failed, the seocndary took over without even twitching. How cool is that!
Hidden “feature”: Split datastores!
A very cool feature I saw but was not noticed by many people (I think) was the fact that the prototype used separate datastores for the primary and secondary VMs. Wait for a second, go back and read this again: Both primary and secondary VMs were using different datastores.
That may not seem like much, but in fact I found this VERY exciting: The fact that both primary and secondary obviously both read but also both WRITE to the datastores means that they basically have synchronous replication going on between these two datastores! Wouldn’t it be the greatest if you could put each side on a different SAN as well? That would protect the VM for host failure, but also a total SAN failure (if you have the luxury of two SANs).
My Take
These are those little tiny details that make VMworld worthwhile for me. Will we be seeing FT protecting against SAN failure as well? When will the fun ever stop? My guess… I won’t. Just enjoy the ride!!

LinkedIn
Twitter
Nice, What I have been looking for for ages.
A lot of people have. Beware: It is still a protoype. No saying on how or when it will be available. Still…. It’s coming!
Bye-bye, Stratus!
Yep, if you’re in this ahead-of-technology business and are not moving on to even cooler things, you’ll get driven over by technology (Marvin the MonsterVM probably at the wheel).
[…] session, the SMP-version of Fault Tolerance with very cool demo (you can read my blogpost on it: 4 vCPU FT demoed with synchronous replication?). Also some cool sessions on PCoIP optimizations in vSphere / View 5. All cool […]