Networks are becoming more software-centric, with the goal of making communications service providers more agile, cost-efficient, quick to market, and less vendor-dependent. But while software is where much of the value resides, the fact is that it has to run on something, and that something is hardware.
As communications services providers move to implement network functions virtualization, much of the hardware on which virtual network functions will run will be in the form of industry-standard hardware. To many that translates to mean x86 servers. However, in at least some cases, VNFs may be able to benefit from higher performance hardware solutions, according to some people in the networking community.
Metaswitch Networks CTO Martin Taylor talked about this potential in a presentation, and an interview with me, earlier this year. More specifically, he said that ARM (News - Alert)-based processors and intelligent NICS are two examples of technology worth exploring in the NFV realm.
More than 60 billion ARM-based chips have shipped to date, he said, but ARM technology until recently has been focused on use in small, low power devices. Now, however, ARM-based chips like Cavium’s ThunderX_NT system on a chip, which is optimized for NFV applications, are becoming available. This solution has the potential to deliver lower power consumption, higher density, and lower total cost of ownership than does x86, making it a potential challenger to that technology. It may also be attractive to CSPs, Taylor added, given that Intel (News - Alert) is an industry giant, and one of the goals of NFV is to help carriers avoid vendor lock in.
ARM, which Japanese media and technology giant Softbank recently announced plans to purchase, has already been active in the NFV space through its involvement with OPNFV. In November, Enea (News - Alert) announced the Pharos Lab for OPNFV; this is a place where network developers can validate applications on ARM-based servers.
And Metaswitch recently ported its Clearwater IMS solution onto the Cavium solution to see how it would perform.
“It’s the beginning of a journey,” he said.
Intelligent NICs also have the potential to allow for better performance by enabling the VNFs to offload some work to them, Taylor said. That’s important, he explained, because the SDN fabric of NFV at the moment is a bottleneck because it doesn’t perform well enough.
Intel is working on that with DPDK, and that will get us part of the way there, he said, but in doing so it will take up a lot of resources on the host. So purpose-built hardware on the NIC (News - Alert) might be a good option following that, Taylor suggested.
However, he said, the telcos may require some convincing that this is mainstream and standard, and not an encroachment of specialized hardware into their new networks. But, he added, some of these NICs are showing up in hyperscale data centers.
“NFV is about leveraging commodity hardware,” he said. “The more interesting question is: What is commodity hardware?”
Edited by Alicia Young