Software defined networking - it's Intel inside.

The recent rapid development of software defined networking and network functions virtualisation may owe more than you might think to the world's leading chipmaker.

On several occasions I've talked to vendors of firewalls, intrusion detection devices, applications controllers etc who've told me they have virtualised a product that was once dedicated hardware so that it is now software running in a virtualised environment.

I've asked them how they have managed to achieve comparable or adequate performance from such a configuration and never received an explanation, only reassurances. Now Brocade subsidiary, Vyatta, has provided one: it's thanks to the increased packet processing power of Intel chips.

Vyatta (the name comes from Sanskrit, it means 'open') hit the market in 2005 with the launch of a software router based on standard x86 hardware. The company was acquired by Brocade in 2012 and now forms the core of Brocade's software defined networking technology. Former Vyatta executives Scott Clark - now senior director, worldwide business development, software networking at Brocade - and EJ Dath - senior software engineer, worldwide business development, software networking - briefed journalists this week in Sydney on the Vyatta platform and Brocade's view of software defined networking.

According to Clark, "Intel has continued to enhance its chipsets and the performance has got better and better. We saw a tipping point in about 2010 when there was enough throughput that a virtualised machine could adequately do routing, firewall and some of these network services on an x86 box along with the applications themselves."

He observes that this coincided with the terms software defined networking and network functions virtualisation starting to gain traction "because now you had enough performance and throughput and capability to shift from proprietary systems to open x86 systems."

Clark conceded that custom silicon would still hold the edge over standard chipsets but said, "With the application of Moore's law you get to the stage where it's good enough and that is where we are right now. A lot of vendors, including Brocade are now starting to ask what are we going to continue to build on ASIC and what can do by leveraging x86?"

According to Clark Intel then looked at what it could do to make its generalised processor more suitable for packet processing, and came up with its Data Plane Developers Kit (DPDK). "Through utilising special command calls you can take advantage of these special opportunities within the Intel chipset to push packets faster," Clark explained.

He said that, prior to exploiting DPDK, Brocade had been able to achieve throughput of 1.9gbps per CPU core, but no more. "Even though you could add more cores the performance was not changing because of the way it had to be architected."

Then DPDK came along. "We were one of the first companies to get access to DPDK and we started to write our software to take advantage of it and now we are getting 10gig throughput per CPU core."

According to Wikipedia, DPDK is "a set of data plane libraries and network interface controller drivers for fast packet processing on Intel Architecture platforms." It "provides a programming framework that scales from Intel Atom processors to Intel Xeon processors and enables faster development of high speed data packet networking applications." It is provided and supported under the open source BSD license.

CONCLUSION

The continued evolution of software defined networking and network functions virtualisation is clearly dependent on semiconductor development continuing to obey Moore's Law. According to Clark, there are no signs from Intel that this will not be the case.

Intel, of course, would be the last to sound the alarm on this one, but others are already seeing portents. In June Wired magazine published an article  Is the End ofMoore’s Law Slowing the World’s Supercomputing Race? It was about a list of the most powerful computers on the planet, compiled every six months by a team of supercomputing academics. Wired reported little change in the list in the past 12 months and quoted Jack Dongarra, a computer science professor with the University of Tennessee who has long been involved with the list, observing: “Things seem to be slowing down. You might characterise it as maybe a sign that Moore’s Law is having some issues.”

Is this list an early warning? Intel of course says no. The article quoted a company spokesman saying “We expect Moore’s Law to continue to provide benefits for the foreseeable future across a broad range of computing segments."

Time will tell.