In the same way that T-Mobile sought to break the mold of telco behavior by proclaiming itself an “Uncarrier,” NFV appears to be having a similar effect on telco network infrastructure. In fact, so much so, that it wouldn’t be far out to talk about the “unhardware” telco in some circles, given some of the hype around the “software” telco vision.
Now, don’t get me wrong; no one’s saying that hardware isn’t needed at all, unless maybe if the carrier is an MVNO. However, the NFV initiative has created very high expectations that an increasing number of telco network functions will be able to run on “bulk standard” data center servers – the sort that telcos already have running their IT software. There’s no problem with that, but I think we are beginning to see an inevitable dose of reality creeping in (see here and here) and we’ll likely see more and more of it as NFV moves from PoCs to live trials this year.
It’s not a big ask to accept that control and signaling plane functionality such as the IMS core can run quite happily on these data center servers, given enough thought has been given to provisioning, redundancy and backup to guarantee the same five-nines availability. However, when it comes to packet and media processing in the data plane, that’s another thing all together and the jury is still out on this, particularly around vSBCs. That said, stellar work is being done in this area too, with SR-IOV and Intel’s DPDK appearing to solve many latency issues. But, as to being able to scale to meet the needs of any reasonably sized operator, NFV is far from a done deal. Conversations that Current Analysis has had around vEPC PoCs reveal that they are frequently small scale and often far from data plane-intensive (like running machine-to-machine applications).
Simply put, they are what they say they are – proofs of concepts – and little more.
So, while the concept of hosting VNFs on “bulk standard” data center servers is still being proven out, what are the prospects for live NFV deployments at scale? Well, most players agree that more work needs to be done, and seemingly right on cue, there was apparently a demonstration of a virtual network platform as a service (VNPaaS) yesterday in Okinawa by NTT, Fujitsu and Alcatel-Lucent Japan. In short, VNPaaS, the thirteenth ETSI ISG NFV PoC, aims to take the reliability and scalability issues away from VNFs and put them into the servers.
Yes, you read that right… into the servers.
It looks like we may need a new generation of server, or at least a new server architecture, one which is optimized for NFV. Now, is this progress or are we actually trending back towards telco-dedicated hardware configurations? Maybe this is the proverbial “two steps forward and one step backward” that marks most successful endeavors. The optimist in me would like to think so. Still, I can’t help but feel some mild déjà vu as I recall the hard-fought steps the industry took towards getting to its current hardware virtualization platform, ATCA, over a decade ago. ATCA showed that the telecoms industry already knows how to virtualize network functions; it’s still the hardware platform which remains the bottleneck. What we need now is real innovation in middleware if we are really going to fulfill the NFV dream and come anywhere near to an “unhardware” telco.