George Symons, Gridstore
Gridstore as a storage company is obviously focussing on SDS. Virtualization has changed the way storage vendors need to look at their storage solutions, because their storage now needs to react on how applications work to be able to provide an optimal performance. And of course cost. How customers want their storage solutions is about cost. It needs to be cheaper, perform better and provide more insight in what the data is actually doing there.
Virtualization changes everything and nothing
In many ways storage always sees what it was always seeing up above: servers doing I/O. Whether these servers are hypervisor hosts with VMs running on top of them or simply physical machines doing their own thing, to the storage it’s like nothing has changed.
The other way around is the same thing: physical machines as well as virtual ones still see a place to land “down there” and if it works, it works. What else is new?
I/O is randomized
But…. when looking a bit deeper at what Hypervisors are doing it’s now like a big randomizer machine, since a lot of different I/O profiles come together and everything is randomized.
THE big issue nowadays is application performance, but the high cost of storage is a big concern too.
Since the era of virtualization that old familiar LUN is now supporting a large number of virtual machines and no longer one specific application and so it is more difficult to have a specialized LUN for a certain I/O profile. Add replication and other higher layers of storage functionality and you can imagine that this all isn’t contributing to ease of management. The LUN is not tuned anymore the way it was before.
Turn it upside down please!
If the control plane is moved up to the VM itself as well as the data plane, this reveals the idea of giving the server more control over the underlying storage. In fact if you go up high enough, you end up giving the application the right amount of resources it needs to function well.
Moving the control up to the server, means you’re using software defined storage so the server defines the storage. The server will manage the replication, manage the way data is laid out over the back-end. This way I/O is optimized in a per application way. The storage could never do this kind of management.
Ease of management can work for you
Storage is the bottleneck, not the VMs or the applications. Separate the control from the data. Gridstore offers a Hyper-V integratable software “driver” which can do a sort of quality of service per VM. They call it managing, but if the storage is already the bottleneck and applications suffer from that, creating a sublayer between Hyper-V and the VMs, this is indeed something like QoS and you need that to rule out the disturbance “noisy neighbors” can cause. But the thing is that you need to install a driver inside the VM to make it application aware. Is that what we want? If the Hypervisor was VM or even application aware and you didn’t have to install something inside the VM, that would really be great, since then the VM can simply keep on doing what they are doing now and the underlying layer has the intelligence to streamline I/Os and make sure performance is how you want it.