luchschen_shutter - Fotolia
One of the key challenges SDN faces going forward is keeping up with the tremendous and always-increasing demands for traffic processing and forwarding. The answer is likely to come from a rapidly emerging offshoot of supercomputing known as inverse virtualization -- the use of many processors to address a single, large and very processor-intensive application.
Just about everyone involved in IT is familiar with the concept of virtualization, at least as applied to computers today. Run a hypervisor, create multiple virtual machines, run multiple operating systems isolated from one another, and make better use of the enormous power of today's microprocessors. Everyone goes home happy, from end users to system administrators to CFOs.
This strategy works really well for applications that demand less than a processor can really deliver. But a huge number of computing problems demand not just the power of one microprocessor but of many, which is where emerging inverse virtualization enters the scene. Now, with SDN continuing to gain traction in data centers worldwide, inverse virtualization could play a bigger role in the years to come in enabling high-speed networks.
Inverse virtualization has its roots in supercomputing, which harnesses thousands of microprocessors ganged together -- often in clever ways -- to make them look like a single, very fast processor suitable for addressing intense traffic demands. The idea is to make the multiple processors appear as one to an operating system or a given application -- a so-called single system image. Techniques that hide, or abstract, the individual processors -- and the underlying complexity -- are another element of inverse virtualization, as are very high-speed interconnections that enable sharing processor memory (RAM) across high-performance busses.
Dynamic SDN to benefit high-speed networks
The bottom line is that we can build very fast computing environments by integrating multiple cores and the processors embodying these cores via high-speed network interconnects and fast memories to address even specialized problems. In the past, these problems were either unsolvable or addressable only with specialized hardware. Interestingly, this hardware can include routers and high-speed Ethernet switches -- the basic building blocks of today's networks. At the same time, it takes even greater advantage of the benefits of SDN.
So, it's no surprise that the inverse-virtualization approach is seeing increasing application in network equipment, like switches and routers -- with this interest and activity being driven by the rapid evolution of SDN. After all, what's a switch or router today? Just fast computer processors with specialized copper or fiber I/O interfaces.
SDN demands that the processing that occurs on these boxes be more flexible and dynamic than in the past, with more functionality implemented in the software. At the same time, however, the fundamental requirements for speed and capacity must also be addressed, and this is where inverse virtualization really shines when applied to high-speed networking. The resulting solutions can be inexpensive, cost-effective, scalable and absolutely in sync with the evolution of SDN itself.
Consider functions such as the SDN controller. It can be virtualized entirely, winding up in a virtual machine or even in the cloud via network functions virtualization, or integrated into SDN-based switches and routers. Ditto for the essentials of analytics and performance optimization -- even though these can be computationally intensive, their requirements are easily addressed via the inverse-virtualization approach, as is the case with basic traffic handling. So, while we'll always need boxes that the various wires and cables essential to networking plug into, the internal architecture and structure of these core networking elements will make increasing use of inverse virtualization. This will also speed the evolution of SDN while addressing the ever-increasing demands in the performance, capacity and scalability of high-speed networks.
Network virtualization re-evaluated after company discovers platform bug
Hybrid approach could work best for network virtualization efforts
The network hardware you'll need for virtualized networks and SDN