Would you put a Virtual SAN in a brownfield? Or a minefield?
Coexistence is the hardest part
Sysadmin Blog In my previous article I looked at the various virtual SANs on the market. In this article, I'm going to take a look at the issues they face trying to get customers to migrate from existing storage solutions. Brownfield conversion strategies are an issue for all players.
If your goal is to get rid of SANs by distributing storage throughout the datacenter then a few issues exist. Existing storage providers provide peace of mind that converged infrastructure companies are going to find hard to overcome. Everyone wants someone else to be the first through the minefield.
Existing SANs also represent sunk costs, not just in the SANs themselves, but in training, network infrastructure, network architecture, performance balancing and application design. The higher up the enterprise stack you go, the more each of these means.
An SMB could switch without too much issue, but convincing a company running their Oracle database of the ragged edge of possibility is going to be a far greater challenge.
All points of the virtual compass
The first hurdle to overcome for converged infrastructure players is that a company's switching infrastructure has to handle the new traffic and the new traffic patterns.
In a SAN environment your SANs were "south" of your servers: storage calls were north-south and everything from your wiring to your NIC layout will reflect that. In a virtual SAN environment your traffic is now going east-west: can your design handle the change?
With the exception of large enterprises that still bring in hardware on a per-project basis, conversion is eventually going to require workloads coming off the existing SAN and going onto the virtual SAN. This does not mean that the compute side of these workloads will necessarily move, just the storage. That's an issue.
Companies with a massive investment in still-perfectly-functional compute nodes aren't going to just throw them away in order to use a virtual SAN. That adds yet more east-west traffic to the mix, but it also introduces a latency problem.
The benefits of converged infrastructure largely come from the fact that workloads and the storage that feeds them live on the same physical hardware. Copies of the data are sent to other nodes, and in a pinch a VM can go east-west to get access to its storage if a local copy doesn't exist yet (say after a vMotion) – but by and large a given VM is talking to the flash and rust in its own node.
Not so for any storageless compute nodes hanging on to the virtual SAN from outside. These will have network latency, but also will have to deal with the fact that no matter how well designed the software is, a virtual SAN isn't a real SAN. No individual node is designed to cope with the kind of loading that a real SAN will experience. "External" compute nodes can ultimately upset the storage/compute balance: load even one node in a cluster too heavily and the whole cluster suffers.
Getting into existing shops
Enterprises install a server and then they retire a server. They are buying nodes by the hundreds and thousands and they just don't "upgrade" systems. Sometimes storage and compute are bought together, sometimes not. If you are bringing in a virtual SAN solution into a brownfield environment it is going to have to be able to work with what already exists.
A lot of the deployed compute nodes that you'll see in enterprises were shipped with some amount of flash over the past years simply because this allowed vendors like Cisco and Dell to drive up the margins for something that ultimately wasn't ever designed to actually get used. This is typically a single flash device: useless to anyone, except as a read cache, a circumstance which might well prove useful to vendors of virtual SANs.
VMware has a head start on its competitors here. It has flash caching integrated right into the hypervisor. It needs to be tediously configured on a per-VM basis, and it doesn't dynamically allocate resources based on changing VM workload, occupancy or available cache, but it's there. How the other vendors deal with the brownfield problem will define the space in 2014.
Nutanix and SimpliVity spend a lot of time on managing brownfield conversion messaging. Maxta is still getting its marketing together and VMware is watching to see what everyone else is up to.
Existing storage providers are sensing the threat virtual SANs pose, and everyone from EMC to Tintri is going on the offensive against the virtual upstarts. EMC did so by buying up virtual SAN vendor ScaleIO and its marketing people are aggressive in getting out the message that EMC won't be crippling the technology to protect its traditional technologies.
Networking companies are starting to bring out marketing targeted directly at this growing segment as well. If the networking companies think there's money to be made here it leads me to believe that people with a lot of experience in the industry believe that virtual SANs and the associated converged infrastructure datacenter designs are here to stay.
What remains to be seen is if virtual SANs will be a niche, or whether they can truly displace entrenched network designs. A lot of this will depend on if the virtual SAN vendors can crack the brownfield code. Don't expect 2014 to tell the definitive tale on this, but 2015 promises to be interesting times indeed. ®