• 07/16/2014
    7:00 AM
  • Rating: 
    0 votes
    Vote up!
    Vote down!

Why Facebook Wedge Is Revolutionary

The social networking giant's open TOR switch is a big step forward for white-box switches and a ground-breaking platform for application networking.

Facebook's Open Compute Project (OCP) switch, code-named Wedge, likely won't displace entrenched networking vendors like Cisco. But here's why it will be revolutionary: It addresses the need for application-specific networking.

Wedge uses commodity hardware in the form of a microserver plus an open-source operating system code-named FBOSS. Wedge and FBOSS  give networking teams the flexibility to create custom switching platforms for specific uses, from both the hardware and software aspects. For example, if a rack only needs 24 ports for connected servers, you can use the modular hardware to build that configuration. If the operating system needs modifications to support advanced features like monitoring or custom software, FBOSS is more than capable of being extended with the proper development team.

As a platform, Wedge -- which Facebook unveiled last month -- is an important step forward for commodity switching hardware (commonly called "white box"). It shows that hardware doesn't need to come from an established networking hardware vendor to be useful in the datacenter.

While Wedge does have long-reaching implications for application networking in the datacenter, the idea that it could displace the big networking players is a bit of a stretch. Wedge isn't designed to supplant existing chassis-based units or even core switching functionality like that found in the Nexus 7000 platform. Instead, Wedge is designed to give Facebook more control of the platform at the top-of-rack (ToR) level. While ToR switches do take up a large amount of real estate in the datacenter, it will take a more coordinated effort to displace incumbent vendors.

Wedge is a modular platform that gives Facebook two important features that aren't found in traditional hardware. The first is the modular server-style architecture. Facebook touts the ability to swap in any type of server control module and retain the feature set of the switch. This is a huge advantage for Facebook, which may buy thousands of control units at once, but less important for enterprises that don't normally buy at that scale.

The second feature of Wedge that Facebook relies upon is the monitoring capability afforded by a Linux-based operating system. Rather than attempting to modify existing vendor software code to work with Facebook's extensive monitoring system, it was much easier to write a switch OS that was compatible and make it control OCP hardware. With FBOSS, Facebook can monitor the network hardware at a deep level to keep its systems running at peak performance at all times.

Shops that are looking at extensive modifications to current vendor switch products are going to benefit the most from Wedge and FBOSS. Their teams are already set up to support the modifications done to bring vendor hardware up to their specifications. With FBOSS, those customizations will be in the OS itself, not added on to work around issues. That means support will largely be an in-house offering.

Customers not looking for the deep application-specific capabilities of FBOSS can still benefit from Wedge in very unique cases. For instance, in situations where a networking appliance is needed to monitor the flow of traffic at the top-of-rack or where data must be manipulated before being passed on, FBOSS and Wedge would make a great fit.

The possibility also exists for application vendors to use Wedge and FBOSS to their advantage. Where previously a server appliance was used to create network collectors and controllers, Wedge can now take its place. A vendor need only take the OCP reference platform and reuse it with proprietary extensions on top of FBOSS. This gives the appliance the power of a server with the modularity and extensibility of a network switch. With the ability to modify the operating system to whatever needs have to be met, a Wedge network appliance could be customized to suit any customer need.

Wedge and FBOSS on their own are not going to change the networking world. They are just OCP reference ideas that ingenious developers and engineers can build upon to transform the network from a passive transport to an active participant in the application ecosystem. Wedge is a bold step in positioning white-box switching in the new application economy.


Rose by any other name

Agreed it's sweet tech, but they couldn't do better than "Wedge"? 

Re: Rose by any other name

Ha! Yeah, I don't get how they were inspired to name it Wedge. Ugh.

Re: Rose by any other name

I assumed it was a fondness for blue cheese. Or "Balls Of Fury."

Re: Rose by any other name

Ah, yes, Wedge McDonald, right?



Open Compute Project

Lots of good things can come out of Facebook's OCP, and some good things have already been created and shared with everyone, for example, power management and power utilization designs that are efficient, etc.

Switch modifications is another area that could prove to be revolutionary in datacenters of social media platforms -- datacenters that need to respond to user input or remote machine input. While, datacenters that act like a host rely more heavily on computational resources.

Re: Open Compute Project

I agree Brian that OCP is producing a lot of promising work, but it still seems that much of it requires an organization with enough resources to take advantage of it. 

Re: Open Compute Project
Marcia, that's my thought as well. I'd really like to see an organization take some of this advanced technology and package it so that your average networking team could use it. I guess we are a few years away from that yet. I know some of the smaller SDN vendors are working on it, but it still seems pretty complicated.
Re: Open Compute Project

Hi Susan, glad to see you thinking some of the same things we were when we first heard about Facebook Wedge here at Pluribus Networks. You see, we have been building and shipping server-switches and network computing appliances with fused network, compute and storage for years. Early on it became clear that there was an opportunity to do interesting things with highly programmable networks, but while much of the industry zigged with central controllers talking to switches with relatively little CPU, we took a different approach and put server class Xeons in the same box with Intel and Broadcom ToR switching and hooked them up via high speed PCIe and integrate Ethernet. With this approach, not only is the CPU big enough to do useful things, the connection between it and the switch is fast enough to enable some really interesting applications in areas like analytics, cloud and even financials. Of course, if you are going to package a server class CPU, you might as well build a real Network Operating System with server style clustering and bare metal hypervisor capabilities, which is what we did with Netvisor. Now, instead of dumb pipes, you have applications running on smart pipes and even better, you can have your applications use standard APIs to get hooks into those smart pipes so now instead of an application aware network you can also have a network aware application. Really the convergence of netops and devops in both hardware and software.

Re: Open Compute Project

Hi Jason, thanks so much for chining in and giving us the details on Pluribus. That sounds like a great approach and I'm interested in learning more. Does Netvisor allow administrators to connect and manage several of these boxes at once? (To me, that seems to be the challenge of other "packaged" solutions -- you lose the resource-sharing and end-to-end agility that are part of the promise of virtualized networks)

Re: Open Compute Project

Hi Susan, one of features of our *nix-based OS, Netvisor, is exactly that. We call it a "fabric-cluster" - perhaps a good way of imagining it would be to picture server type clustering, three phase commit and all, coming to the network. Add a new box, add it to the fabric-cluster. Whole fabric is administered as a single logical entity from any box in the fabric and all have visibility into the network and traffic on the network. The whole idea is to simplify and consolidate, removing complexity rather than adding overlays or additional networks for monitoring etc. 

Re: Open Compute Project

Gotcha. I read about fabric clustering on your website, but I wanted to make sure I was interpreting it correctly. Sounds like great technology -- thanks for all the info.

Re: Open Compute Project

@Marcia, I completely agree and feel it is a time equation.

I tend to view the OCP as an outcome of monopsony, as Facebook's requirements for a datacenter is quite unique. Facebook grew to scales at which they could customize their needs and suppliers would happily build the customized product.

Since, data is changing into big-data and connectivity is expected to increase exponentially, I feel we will soon see other firms utilizing these Open designs.

Re: Open Compute Project

I guess introduction to open network switch was another leap in OCP and this shall certainly help other firms as well.