It’s been said that software defined networking (SDN) is server virtualization on the network. It isn’t really, but the effect is largely the same, and it does make it easier to marry dynamic virtual instances to the proper network resources.
The thing that many people forget is that server virtualization was not the slam-dunk technology it’s been made out to be. Sure, it greatly improved server utilization to reduce the data center hardware spend, but much of those savings had to go into the storage budget to provide the capacity all those virtual machines demand. For anyone who’s ever wondered why EMC took a controlling interest in VMware, that’s the reason.
So that raises the question: are there unforeseen consequences to SDN, particularly in storage, that will only become widely apparent after deployment is underway? It seems that there are. But there are also a number of fixes in the works that should help alleviate any imbalances before serious damage is done.
The problem is that all that activity coming from abstract resources can quickly overwhelm anything still operating on the physical plane alone. The obvious solution for storage, according to the SNIA’s John Martin, is to introduce a virtual layer as well. The way to do this is to create a software layer the can manage data apart from the actual infrastructure responsible for storage, copying and retrieval. Once this is done, managers will have an easier time leveraging their various traditional disk, solid state or advanced memory solution storage tiers.
It appears that the enterprise won’t have to wait long for such a scheme to arrive. Companies like Coho Data are already pushing advanced scale-out platforms into the channel that provide modular, flash-based solutions capable of matching virtual and SDN loads bit for bit. The company’s DataStream 1000, for example, utilizes open commodity hardware, automated tiering and advanced analytics to accommodate dynamic data loads, doing away with much of the manual configuration that makes traditional storage the slowpoke of the enterprise.
Coho has even managed to put an SDN switch directly on the appliance, says CTO Andy Warfield. This allows data to avoid the storage bus entirely and connect to the network via the PCIe interface. This then allows much of the storage addressing and presentation logic to be pushed onto the network, where it can be more closely aligned with application or user needs. At the same time, it provides advantages when it comes to the limitations of single-IP endpoints in protocols like NFS and the need for high-performance, multitenant isolation in virtual environments.
HP is in the channel with an SDN-optimized storage solution as well. The company’s StoreVirtual 4335 is a 1U device that leverages the company’s Adaptive Optimization system to auto-tier across traditional and solid state drives. The system keeps track of data access patterns and other metrics to ensure high-speed storage functionality for the most critical data. It also drives up storage utilization by allocating replication and parity data on lower-performance SAS drives. The device is overseen by the LeftHand Operating System (the former SAN/iQ) that provides support for thin provisioning, smart cloning, snapshots and other functions.
The takeaway here is that SDN will not deliver the fully virtualized data environment on its own. In fact, deployment will have to go hand-in-hand with the development of server, storage and mobile infrastructure if the enterprise truly desires a fully abstracted data environment.
Magic bullet solutions are rare and wonderful things, but not even SDN can deliver dynamic data operations if the rest of the data center isn’t ready for them.
Photo courtesy of Shutterstock.