Software Defined Networking (SDN) has always looked a bit like a solution in search of a problem, at least in the enterprise data center. But there are lots of potential applications that need a dynamic and scalable network. In my mind, storage is chief among these, since scalability and flexibility has always been extremely difficult to achieve.
As I have written before, one of the most vexing problems in storage is the backwardness and intractability of the protocols for we use. SCSI and NFS simply were not designed for flexibility and scalability, yet they are the primary protocols in use in today’s data center.
Numerous attempts have been made to overcome these limitations, but few have gained any real traction. Where is pNFS? Where is FCoE? Even iSCSI has been relegated to smaller businesses and applications. The only real success in evolving storage protocols for the modern world is Microsoft’s SMB, and that has more to do with Redmond’s ability to push through changes then any real customer demand.
There have long been just three ways to add flexibility to storage:
- Change the protocol, switching to something more dynamic like XAM, pNFS, CDMI, etc. But this requires client and server adoption, and one or both of these have derailed each of these efforts. The only real success so far is Amazon’s S3 protocol, and that’s because developers were so fed up with enterprise IT that they simply walked away and wrote their own protocol! And Microsoft is making real advancements in SMB, if anyone pays attention.
- Trick the protocol using client or server wizardry to address multiple targets, allow live modification to storage allocation, and the like. This is how Isilon can scale NFS like no one else, but it’s a difficult game to play. I’ve seen dozens of storage companies whose claim to fame is some way to trick SCSI into being more dynamic without the client or server knowing what’s going on, and none have made much impact.
- Move up the stack, adding a volume manager or similar host-side storage layer. This is the VSAN approach, which moves storage allocation and management out of the hands of storage devices while using conventional protocols only for low-level connectivity. It’s worked for decades, from Veritas Volume Manager to Microsoft Storage Spaces, and is the core IP of high-profile newcomers from Nutanix to PernixData.
But what if there was a way to get the network involved? Storage has always treated the network like a dumb pipe, pushing bits from one end to the other with a priority on performance and reliability. Although networks could offer high availability, flexibility, and scale, storage has largely ignored these capabilities.
Software-Defined Networking (SDN) could change this picture. SDN is coming to the datacenter, with pretty much every networking provider adopting the concept and specific implementations, like OpenFlow and OpenDaylight. Eventually, the storage industry will “open its eyes” and realize that this capability has huge potential as a “fourth way” to add flexibility to storage.
The most amazing outcome might be “storage area networks” that finally live up to the promise of that term. To date, Fibre Channel SANs have been quite simple in execution, acting as a bus to push bits from one place to another. But SDN-powered SANs could be something else entirely, bringing dynamic, scale-out attachment and adding storage services in the network, not just at the end points.
If you happen to be in New York on May 5, I suggest you consider stopping in for a panel I’ll be leading on this very topic at the Open Networking User Group (ONUG). My panel, “IP Storage Panel: SDN’s Killer App”, is just one of a number of great topics to be covered during ONUG. Use code “TFD10” to get a 10% discount when signing up!