VMware Fault Tolerance overview and limitations

vmware logo

Last month at VMworld 2008 in Las Vegas, VMware formally introduced the upcoming high availability feature simply called Fault Tolerance or FT.

Previewed for the first time in late 2007 with the tentative name of continuous availability, VMware finally shared more details about this groundbreaking capability and promised that it will be part of the upcoming VMware Infrastructure 4, currently in private beta.

Most information were disclosed during a conference session (BC2621: Fault-Tolerant VMs in VI: Operations and Best Practices) which Scott Lowe live blogged here.

Besides a clear description of how FT works, some interesting limitations and requirements emerge from his report:

…VMware FT requires CPUs that support hardware virtualization (AMD-V, Intel VT). These features sometimes need to be enabled in the BIOS of the server. All hosts must be running the same build of VMware ESX, shared storage is required (NAS or SAN), and all hosts must be in an HA-enabled cluster. In addition, a separate FT logging NIC and a separate VMotion NIC are required. This means a minimum of 4 NICs are necessary (Service Console, VM traffic, VMotion, FT logging). Gigabit Ethernet is required for the FT logging NIC (just like the VMotion NIC).

VMware FT can’t protect VMs that are using thin provisioned disks; disks must be “thick.” Disks will be automatically made thick when VMware FT is enabled. VMs can’t have any non-replayable devices (USB, sounds, physical CD-ROM, physical floppy, physical-mode RDMs) and paravirtualization-based VMs are not supported…

Considering the above requirements and current feature prices, it sounds like VMware FT is going to be a very expensive add-on. If so VMware may miss a big opportunity: this feature alone makes virtualization worthwhile of consideration (no matter if you aim at server consolidation or not) and every potential customer looking at ESX may want it much more than any other option.