The key thing for software defined networking is open standards and commodity hardware, says Vyatta’s Kelly Herrell
Brocade has announced it intends to buy software defined networking company Vyatta to strengthen its play in the field, which allows data centres and cloud providers to set up networks more flexibly.
Earlier this year, before the deal was announced, we spoke to CEO Kelly Herrell about Vyatta’s aims, and its transformation from an open source networking company, to a software defined player.
‘We unmasked x86 as a wicked fast networker’
Being a software company in the network space makes Vyatta a rare beast. When this reporter first heard of the company it was an open source company(old ZDNet story via Wayback) with a mission to radically undercut Cisco and the like (another old ZDnet story). But given the conservatism of network buyers, this was an uphill struggle, Herrell told us.
“The open source story was important for us in our early penetration, but is decreasing in importance going forward,” he said.
The company shifted to a message of using commodity hardware instead of the specialised equipment built on custom ASICs, as used by most network hardware companies. “The one thing that we’d always said was that x86 would become a good networker,” said Herrell. “They said prove it, and we have.”
As Intel continued to increase the performance of its chips, x86-based systems became a better bet for managing networks, said Herrell. “There’s a surplus of horsepower that comes in low cost server economics.”
While custom silicon is still best for switches, commodity hardware is easily good enough for jobs like routing and security, he said, “and the routing and security market is bigger than the switch market.”
The opportunity was “shocking” he said: “If you need a 10G router, the minimum cost on proprietary hardware is $100,000, but you can do that on a $1,000 server in our world. And if you needed 40G you can can scale the performance linearly.”
The arrival of software defined networking lent credibility to this, and the OpenFlow standard made a small player look more credible – but we can imagine the move into Brocade will help Vyatta open doors to customers.
Herrell discussed the possibility of using software defined networks within individual virtual servers, driven by the increasing power within those servers.
As servers get upgraded, the new ones have more and more cores, and support more and more virtual servers, he said. But this brings a problem he described as “clotting up the network” in that many virtual servers on a single server presents a new layer of abstraction, which conventional network kit can’t see into.
Herrell’s idea is to put a software defined router on each server. “As traffic pours into that server, the first destination is that router.” he said. The software router on that server then passes traffic to virtual machines, according to firewall rules and router paths, which exactly mimic the network architecture between real servers, and means traffic doesn’t have to go outside the server to get routed back to another VM on the same server.
The process extends network capability into the server, where other approaches such as VMware’s VMotion “break” the network, he claimed. The sheer number of virtual machines in a data centre can be “mayhem” for traffic patterns. “Virtualisation radically alters layer 3 traffic patterns, and we need a new solution to get it under control,” said Herrell. “Otherwise traffic must keep going out of the box and back in.”
The fact that the traffic is handled with normal network tags and protocols within the virtual router is a plus he said: “When traffic has to lave that server it will play just fine on the external network.”
He’d like to see every virtualised server fitted with a Vyatta router of course. “The world would be a better place if that happened.”
Standards are the future
As a small player in networking, Vyatta had to be agnostic and meet standards. The router runs on all hypervisors, he told us.
For the future, SDN will have to be based on standards – and in particular OpenFlow, which looks at networks in terms of the control plane and forwarding plane – or “pitcher and catcher” as Herrell calls them. OpenFlow separates the control plane (which manages the traffic) from the forwarding plane (which actually sends the traffic)
“The first thing we are talking about is distributed forwarding, which sets up a centralised controller talking to forwarding planes all over the data centre,” he said. “It separates the network device into parts and deploys them all over the data centre.” Vyatta does this with something called vPlane.
Anyone buying into SDN should make sure their vendor follows standards – at both the control plane and forwarding plane: “Some vendors have the pitcher and catcher under their own control.”
The idea could result in big “catcher” forwarding plane switches on the top of server racks, controlled by OpenFlow “pitchers” on commodity hardware in the server room.
Hewlett-Packard, Brocade and others have all given verbal support for OpenFlow, but Cisco, predictably, does not support it, preferring its own UCS single-vendor virtualisation solution.
“Cisco has said ‘over our dead body’.” said Herrell. “For this to work, the technology has to have a full ecosystem vendors – and everyone else has said they will support it.”
What happens when tech goes to the movies? Try our quiz!