NFV and How You Will Evaluate the Packet Core
April 14, 2014 Leave a comment
We spend a lot of time helping vendors with their messaging. You’d think, with years of practice behind them, they wouldn’t need much help. The reality is, particularly with new technologies, the messaging can be tricky. How, for instance, can you talk about technology roadmaps and product improvements without implicitly admitting that current products and solutions are still in need of some work?
Yep, it’s a fine line.
This came up recently on a briefing around network functions virtualization (NFV) where a vendor spent some time touting a partnership with Intel. The vendor claimed that work with Intel’s Data Plane Development Kit (DPDK) was going to support the data plane optimization critical for NFV applications such as EPC when moving to server-based platforms – you know, applications where the packet processing performance is a major concern.
On our end, there were two implications.
- If packet processing improvements are coming, then the suggestion is that today’s virtual EPC (vEPC) implementations aren’t ready for prime time. At the very least, they’re not as robust as they could be… or will be in the near term.
- As we move from EPC solutions built on dedicated hardware such as ATCA and proprietary routing platforms to virtual solutions built on common hardware platforms, differentiating based on things such as packet processing performance will become more and more difficult.
There’s another way to think about number two. If every vendor were to use the same server platform for their vEPC implementations, they would all be subject to the same, basic performance constraints. However; just as in the case of NASCAR racing, it’s often the tuning and precision by which the various components of the solution interwork and perform that differentiates vendor solutions.
For service providers procuring mobile packet core solutions, there’s also a clear implication. Going forward, comparing mobile packet core solutions based on hardware performance metrics will be much more difficult. What’s the solution then for carriers? Just as vendors need to walk a fine line in how they message product evolutions, operators need to talk a fine line in terms of their sourcing decisions.
Where hardware performance becomes less of a competitive differentiator, operators will need to look to larger solution criteria such as service support, feature support, customer references, and vendor plans for go-forward improvements. At the same time, it’s wrong to position software and hardware as completely divorced from one another. If software innovations (such as those promised by Intel’s DPDK) can improve one vendor’s product performance, then differentiation may result.
Ultimately, the procurement process remains the same as it ever was. Procurement decisions are a result of diverse variables: hardware performance, software performance, software features, support, vendor viability, etc. Virtualization complicates the interplay between these. Luckily, the basics of the process remain the same.