At Goldman Sachs SDN is déjà vu all over again

Earlier programmability at Goldman Sachs was a Rube Goldberg machine

Goldman Sachs has been doing SDNs for a long time. It just wasn't called SDNs when the investment giant invested in network programmability. It was just a bunch of APIs, software development kits and other code used to cobble together a large number of various specialized networks – trading, investment banking and the like -- across the globe.

But it wasn’t an ideal solution.

+MORE ON NETWORK WORLD: SDNs move from theory to reality +

“We’ve been doing software-defined for a long time but it’s all done through ‘Rube Goldberg’ machines,” said Matthew Liste, managing director/technology fellow at Goldman Sachs, referring to the over-engineered machine that performs a simple task in a complicated way. “It was not an ideal way of driving forward in a software-defined environment.”

That’s why Goldman Sachs is eager for software definition to become SDN: the promise of coalescence or uniformity among the various ways to software-define a networking environment.

And Goldman Sachs wants SDN as a way to gain leverage in how its networks develop, Liste said during an address at the Open Networking Summit.

“We really desire commodity scale architectures, software-defined everything,” Liste said. “We‘re big believers in open standards and open architectures…so we have a say in how things evolve.”

That includes programmatic and comprehensive control planes, and an ability to improve data planes and control planes independently. And the method to get there must be more cohesive and seamless than the piecemeal techniques of the past.

One was COPS, the policy protocol that attempted to separate the control and data planes of a switch or router, just as OpenFlow is designed to do today, Liste notes.

“15 years later, we’re back,” he said.

Others were GateD and Quagga, early methods of separating software – routing, switching, Layer 4-7 -- from an appliance, much like Network Functions Virtualization (NFV) defines today. And frame relay, ATM, MPLS, IPSec and L2TP were earlier overlay and tunneling techniques to VXLAN, NVGRE and other network virtualization mechanisms in vogue today.

“The time is right for all of these approaches to become real,” Liste said.

And Goldman Sachs has tried them all: vendor systems for overlay, open source for overlay, hardware only OpenFlow, hypervisor and hardware OpenFlow, hardware traffic engineering and OpenFlow, TAP/Matrix switching using OpenFlow, various whitebox solutions with an open operating system, and Open Stack.

“What have we tried? Everything,” Liste said. “We want to make sure we’re informed. No single approach is close to where we want to get to. But we’re confident we’ll get there over time. This is long-term, it is a marathon.”

So Goldman Sachs is intent on continuing to work on a Layer 4-7 NFV model for virtualized network services, merchant silicon and open source, centralized controllers for policy and traffic engineering,  open source management and provisioning, and further testing and use case examination of overlays.

The investment firm would like to see common control planes where vendors can plug in their drivers; bare metal switches with common hardware abstraction; and Linux-like operating systems for switches with common VLAN establishment and configuration methods. It’s confident it can get there through commercial vendors, community open source, and start-ups.

“We think this is fundamentally important,” Liste said. “We are here for the long haul.”

Jim Duffy has been covering technology for over 27 years, 22 at Network World. He also writes The Cisco Connection blog and can be reached on Twitter @Jim_Duffy.

Insider Shootout: Best security tools for small business
Join the discussion
Be the first to comment on this article. Our Commenting Policies