(EN) #VMworld Europe 2011: #BCO2874 vSphere High Availability 5.0 and SMP Fault Tolerance – Technical Overview

0 Flares Twitter 0 Facebook 0 LinkedIn 0 Email -- 0 Flares ×

I attendeed my first VMworld Europe 2011 session this morning.

First part was basically a recap of the new HA, completely rewritten in version 5.0. The speaker, Keith Farkas from VMware, first showed how HA has changed in the last vSphere release: switch from AAH to FDM, no more primary/secondary nodes, storage heartbeat.

 

Second part, presented byJim Chow also from VMware, has been more interesting: SMP on FT, the possibility to have Fault Tolerance on multi-CPU virtual machines. Even if the presentation was about development software, this feature is eveer wanted by VMware users from the first day FT was launched. In fact, the room was fully occupied.

Many users would be more than happy to protect some of their most critical VMs, since it’s easy to setup and transparent to the Guest OS, so it has no limitations and hard configurations to put in place like on clusters.

However, the 1-vCPU limit has always limited its use, together with the lack of snapshots for the Ft-Protected virtual disks.

 

As told by the speaker, to develop this feature in software, like in FT, is a really hard and challenging task, even for a highly talented and large team like VMware engineers. Despite this, some goals has been reached:

– FT Logging has been upgraded to 10 Gbit to handle the traffic produced by multiple CPU

– a new protocol instead of vLockstep builded by the ground up to manage SMP

– no more a shared disk, but two separated and replicated disks (one for the primary VM and the other for the backup VM)

 

This new technology was also previewed, and it has impressed everybody: a linux server with Oracle database and 4 vCPU. At the forced shutdown of the ESXi node hosting the primary VM, , backup VM went on executing the workload.

From he tests they run, FT performances in SMP degrade compared to the same VM without FT protection. Even if this degradation will remain in the production version, I think will be a fair trade for having FT on multi-processors.

 

Very last question at the end of the session: “when we could do snapshots on FT-protected VMs?”.

Answer: “We are working on it”….