Happy Days

Happy Days at Big Switch

Happy DaysThere’s an old song — I’m not sure who wrote it but I know Barbara Streisand sang it — that goes, “Happy days are here again, the skies above are clear again, so let’s sing a song of cheer again, happy days are here again!

I’m sure that describes the mood over at SDN start-up Big Switch these days.  First, arch-rival (at least in the media) Nicira gets taken out at an obscene price by VMWare, who is likely to bury the product deep in the VMWare stack, creating more opportunity for Big Switch.  Then, in February, the company announced it had secured another round of funding from Intel Capital.  Slightly after the announcement, I visited their offices, so that had to raise everyone’s spirits!  Finally, this week the company announced it was trying to accelerate the adoption of OpenFlow-based software-defined networks by releasing an open-source-based “thin” virtual switch, called Switch Light, for commodity or “white box” switches from vendors such as Quanta and Accton.

Switch Light is software that can be deployed on merchant silicon-based white box switches or even on a hypervisor on a server to create a low cost switch.  This would allow Big Switch to sell an end-to-end solution rather than having customers purchase a controller from them and then have to go elsewhere for the switching infrastructure.  Switch Light-based infrastructure will lower hardware costs by giving organizations a low-cost alternative to switches from the big iron vendors.  Switch Light will be supported on Broadcom’s Trident+ and Trident-2 based switches and Linux-based hypervisors, and will eventually be ported to other platforms.

Switch Light is based on open source implementation “Indigo”, which is a subset of the Floodlight project — the community developing an open source-based SDN controller.  I thought the decision to use Indigo was odd given the momentum behind Open VSwitch (OVS) today, but this does give potential partners a credible Indigo-based product now.

I think it’s important to understand which organizations this will appeal to.  I’ve read over and over how this is taking a shot at the legacy vendors’ — most notably Cisco’s — business model, and how the move will drive pricing down.  The appeal of SDNs for the next five years or so, particularly those deployed using white box switches, will be to large financial service firms, big web companies, and perhaps some niche organizations that are heavily network-dependent, like oil and gas.  The rest of the world will likely look to their preferred network vendors for an SDN solution when they’re ready to deploy.  It’s actually for this reason I applaud the move by Extreme Networks, who was mentioned in this release as one of the partners adopting Switch Light.  Extreme has slowly been gaining ground in the data center with the behemoth switch, the BDX8 data center switch.  The Switch Light-based product compliments the current products nicely and enables Extreme to offer its customers an end-to-end solution.

Consider the impact that virtualization had on servers.  Sure, it created a market for commodity servers for web portals and like organizations, but it also gave rise to Cisco’s UCS, a premium priced server, for customers that needed the extra features and functions rather than just a cheap server.  The network market is likely to evolve around parallel paths for the foreseeable future, as we see more innovation in the white box space to continue to drive costs down and increase programmability at the same time as we see the mainstream vendors build bigger and more feature-rich products.

The downside of the Big Switch release is that it likely requires a “rip and replace” strategy, or at least running a parallel network, by the deploying organization.  This is in contrast to Pica8 who is trying to take more of co-existence approach.

Regardless of the approach, though, the combined number of solutions now made available because of vendors like Big Switch and Pica8 in some ways legitimizes the market by giving potential buyers choices they never had before.

Image credit: jhonnyilbello (flickr) – CC-BY-2.0

  • Nils_Swart

    Optimizing scale and cost of switching and packet processing will continue to happen, and it will happen in many places. Whether it’s in core switches, top of rack switches, network interface cards, hypervisor switches or direct memory access: the challenge is not so much the fact that packet processing happens, but instead it’s the difficulty of managing all of that w/o abstraction and associated trust models. Swapping out integrated vendor X for a combination of vendor (P+Q) does not fundamentally solve any of the more expensive problems such as operations or under-utilization. Proper abstraction requires trust in both directions: 1) trust that those who will be responsible to implement to an abstraction will do the right thing, and 2) trust that those who are asking for something in the abstract, are asking for a valid reason.

    Martin Casado put it quite succinctly in an event earlier this year: SDN -moves- complexity but does not remove it.

    Abstractions are much harder to solve, and solving that has to be a goal in and of itself. We believe “affinities” are a good starting point for that.

  • Pingback: Plexxi Pulse – Cisco, Big Switch and the Changing Market | Plexxi()