Is it time to reimagine the NFV infrastructure?

In the run-up to SDN World Congress in The Hague, it is interesting to note that it was at this very show in Dusseldorf 5 years ago that the original NFV whitepaper was first presented. How time flies!

I am sure that many at that 2012 conference would have expected more progress to be made (at least the optimistic part of the audience), but alas, we are still trying to figure out how to deploy NFV and even to justify the business case. No surprise, some of the cynics among you might say; this is, after all, the telecom industry and nothing moves fast in the telecom world!

But, the promise of the NFV whitepaper and enthusiasm that followed was not just about technology and architectures, but a call to action for the industry to change and learn from its cloud computing peers. What was impressive then, and still is now, is the speed and agility at which the cloud computing industry operates. Faced with what might seem insurmountable challenges in dealing with exponential customer and associated data growth, the cloud companies turned to rapid innovation for the answer.

In 2012, SDN and NFV were inspired by the success that cloud companies had achieved at that point in addressing their challenges with innovative solutions based on standard computing platforms. The call to action was to be inspired by what they had achieved, adopt the approach they had taken and apply this to the telecom environment.

Has the time come to go back to the well and be re-inspired?

In the last 5 years, there has been a lot of activity and discussions on how to take the blueprint of 2012 and make it a reality. What we as a community have found is that it is not so simple. Some would claim that it is because telecom has special needs that are different to those in the cloud services area, but others would contend that if cloud service providers can do it, then the telecom community should be able to also. There are a fair share of critics out there, but if we don’t implement SDN and NFV, then what is the alternative? Do nothing?

While trialing and debating the technology and sticking rigidly to the 2012 blueprint, we might have missed the larger point of the exercise; namely making telecom vendors and carriers more agile, not just in their technology, but in their thinking and decision making!

What is interesting to note is that over the last 5 years, cloud computing has continued to advance with new innovations. Cloud service providers have also struggled to achieve the performance they wanted using COTS and virtualization, but instead of persevering with these concepts, they sought new ways to solve the problem. Cloud computing has advanced beyond virtual machines to containers and micro-services and beyond standard computing platforms to FPGA-based computing platforms.

Containers, microservices and intent-based networking models have made the radar and vendors are starting to offer solutions based on these concepts. There are plenty of good discussions already started on these topics, but I want to devote time here to the other major shift that might have gone unnoticed by many unless they are specifically concerned with the NFV infrastructure, namely the shift from standard computing platforms to FPGA-based computing platforms.

The challenges that vendors and their carrier customers have been facing over the last 5 years in implementing a high-performance, yet flexible and cost-efficient NFV infrastructure based on standard computing platforms are not unique. Cloud service providers have also struggled to get the performance, flexibility and cost-efficiency they need from the same standard compute platforms. This has led cloud service providers to investigate a wide range of technologies and solutions that could accelerate performance and reduce costs.

The breakthrough was made by Microsoft Azure in 2015 when they first talked about using FPGA-based SmartNICs to enable a more programmable software defined datacenter[1]. Since then, Microsoft Azure has gone to deploy FPGA technology throughout their datacenter for a wide range of services[2].

Microsoft Azure might have been the first to show how FPGAs could be used to improve performance, but they were quickly followed by others, such as Amazon with FPGA-as-a-Service[3] in November 2016. It was this general move in the cloud community towards FPGA technology that encouraged Intel to purchase Altera, one of the major vendors of FPGA chips for a record $16.7 billion already in 2015.[4] Intel intends to merge FPGA technology with CPU technology to offer the best of both worlds in standard servers of the future. Today, the best way to introduce FPGA technology into the server architecture is as a SmartNIC plug-in card.

The good news is that for telecom equipment vendors, FPGAs are a familiar technology as they have been widely used in appliances for several years. So, unlike new technologies and paradigms emerging from the cloud world, this is at least one that most telecom guys understand already!

The value of FPGAs lies in their reconfigurability and ability to provide a massive parallel computing platform dedicated to a specific task. This is a useful complement to the generic CPU in standard servers, which has the versatility to handle multiple tasks at once. It allows a standard server platform to be enhanced with extra processing capabilities and options simply by adding a FGPA-based SmartNIC. This hardware platform can then be reconfigured on-the-fly to support various workloads providing performance, but also agility and cost-efficiency as the same hardware is used and reused for different purposes.

So, has the time come to reimagine how the NFV infrastructure should be built? The current ideology is based on the 2012 picture of what was happening in the world at that time with standard compute platforms and virtualization. Five years on, the challenges that have been uncovered and the new innovations in containers, microservices, DevOps and intent-based networking will themselves place pressure on the NFV architecture to adapt. The NFV infrastructure will need to respond to these requirements also and just as cloud service providers have found, reconfigurable computing based on FPGA-based SmartNICs, the NFV community can benefit from investigating a similar approach.

[1] https://azure.microsoft.com/da-dk/blog/report-from-open-networking-summit-achieving-hyper-scale-with-software-defined-networking/

[2] https://azure.microsoft.com/en-us/resources/videos/build-2017-inside-the-microsoft-fpga-based-configurable-cloud/

[3] https://techcrunch.com/2016/11/30/aws-announces-fpga-instances-for-its-ec2-cloud-computing-service/

[4] https://newsroom.intel.com/wp-content/uploads/sites/11/2016/03/Intel-to-Acquire-Altera-Press-Release.pdf