Some of my fellow analysts have lumped Virsto into the flash acceleration category along with caching solutions like Proximal Data, Sandisk's Flashsoft and Intel's CAS. While Virsto can use flash to accelerate some storage I/O, it's not primarily a flash acceleration product. In fact, Virsto is a log-based, clustered file system that uses a dedicated log device, which can be a shared SSD, to accelerate virtual machine I/O.
The Virsto log accumulates multiple random writes from different VMs and aggregates them into a smaller number of more-or-less sequential writes to free space in one or another of the back-end storage pools it manages. This aggregation function cleans up the highly random mess that comes from multiple requests in a hypervisor environment, a mess that traditional storage systems have difficulty dealing with.
But at VMworld last year, Virsto demonstrated a version of its software that, instead of a shared SSD, used the PCIe flash cards from EMC’s VFcache product in several host servers to create a distributed log device that protects data written to it by replicating the data across multiple hosts and their associated PCIe flash. This demonstration may be where my fellow analysts got the idea Virsto is a caching solution.
While the write acceleration Virsto gets from using the log device is nice, as far as I'm concerned, Virsto's biggest selling point is the enhanced storage management it provides. Virsto can support multiple back-end storage pools with different performance and protection characteristics, and can write data to each as appropriate.
However, Virsto's biggest advantage is that it provides per-VM, or actually per-virtual-disk, storage management. Virsto installs into a vSphere host as a virtual storage appliance that presents the storage it manages to the vSphere cluster via NFS. This lets the Virsto VSA see each individual VMDK. The log-based, write-in-free-space file system behind the VSA allows Virsto to create per-VM snapshots or read-write replicas of VMs and/or VMDKs with minimal overhead.
When you consider that VMware's current VMFS snapshots are pigs that charge a significant space and performance tax when used, and that the consolidation of multiple VMs into a single data store makes array-based snapshots less useful, Virsto's excellent snapshot capability is a big plus.
Despite the fact that storage hardware giant EMC owns the majority of VMware's stock, the folks at VMware have been promoting the idea of software-defined storage for about two years. They've talked about virtual volumes (vVols), a form of per-virtual-disk micro-LUN that would use a demultiplexor function that sounds a lot like Virsto's log to provide the performance and per-VM snapshot functions Virsto provides today.
Is VMware's acquisition of Virsto an admission that VMFS is behind the times? Does it mean that the hardware vendor participation the vVols concept required is turning out to be harder than VMware hoped, and that Virsto's technology that solves the same problems on any hardware is needed at least in the short term?
Of course, the other question is what will happen to the Hyper-V version of Virsto. Virsto developed its code for Hyper-V before acquiring Evostor, another VMware storage ISV, and combining its own technology with Evostor's to produce the VMware version. I can't imagine VMware selling software to improve Hyper-V storage performance and management, but stranger things have happened.