VMware & Virsto

by Bob Plankers on February 13, 2013 · 7 comments

in Storage,Virtualization

Howard Marks has a great piece on VMware buying Virsto over at Network Computing (link is below):

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.

I saw Virsto for the first time at VMworld 2012, and it looked interesting as something that tries to turn a lot of the random I/O from a virtualization environment back into sequential I/O that arrays can better handle, while adding a ton of modern features missing from vSphere. I’ve been asked a couple of times now about what I think about the recent news, and much of it mirrors what Howard is thinking. I’d add a couple of things.

First, Howard is correct, VMware’s current disk technologies and filesystems are starting to show their age. VMFS has problems with snapshots and performance, a problem I’ve been dealing with in some of my environments. VMDK may not have actual problems itself, but it also doesn’t help solve problems with I/O alignment, space efficiency/UNMAP, etc. VMDK is getting replaced with Space Efficient Sparse Disks, a new technology rolled out with vSphere 5.1. SE Sparse disks have dynamic block sizes and the ability to reclaim space from a guest OS, essentially re-thin provisioning a VM automatically. While SE Sparse disks are only for VDI right now I wouldn’t be surprised to see it available for regular guests in the next major vSphere release. On the VMFS front VMware has talked about Virtual Volumes, or vVols, which mashes up the concept of the datastore with VASA array communications. Under vVols, Storage Profiles get a lot more powerful because now you can assign performance and availability requirements to a VM, and then the whole virtualization stack from the storage on up can work to manage those expectations. The problem with vVols is that it relies on hardware vendors, which goes against the software-defined data center philosophy that VMware has. It’s also yet-another-kludge on top of other kludges that work around the fact that the way we are forced to use storage hasn’t fundamentally changed since SCSI appeared in 1981.

The second piece of this in my mind is VMware’s Distributed Storage, or vSAN as it’s been called. VMware talked about their vSAN distributed storage work at VMworld 2012, and it’s a technology that would basically turn local storage into replicated, usable space for hosting VMs. It would turn VMware into a competitor in the Nutanix/Simplivity/Scale space, except with hardware agnosticism. You take your run-of-the-mill Dell PowerEdge R720, cram some SSD and SAS in it, add it to the cluster, and the cluster manages the rest. Just like building blocks for the data center. For VMware is that this is brand new technology they are building from the ground up. That’s a problem, since it takes a lot of time & effort to come up with those new technologies, and even more time to make them reliable. Nutanix just shipped its third major release of their products (NOS 3.0) and definitely has first-mover advantage in the converged, building-block space. VMware doesn’t even have a 1.0 product, and with VMware’s reputation for QA and the fact that a second major release would be two to three years away VMware is basically out of this game.

So you see why they bought Virsto. First, Virsto has a working, shipping, proven code base right now, just needs integrating. Second, the product fits directly into the software-defined data center vision, bringing the logic back up out of the arrays and into the virtualization layer. Third, it brings to the table a load of modern storage features, like log-structured filesystem techniques, replication, use of SSD and flash for acceleration, storage tiering, thin provisioning, better snapshots, etc. These are all things VMware doesn’t have to invent anymore for VMFS, VMDK, vVols, and vSAN, they just have to integrate Virsto and roll the product out the door.




Comments on this entry are closed.

{ 1 trackback }

Previous post:

Next post: