How To Build a 100,000-Port Ethernet Switch

BobB-nw writes "University of California at San Diego researchers Tuesday are presenting a paper (PDF) describing software that they say could make data center networks massively scalable. The researchers say their PortLand software will enable Layer 2 data center network fabrics scalable to 100,000 ports and beyond; they have a prototype running at the school's Department of Computer Science and Engineering's Jacobs School of Engineering. 'With PortLand, we came up with a set of algorithms and protocols that combine the best of layer 2 and layer 3 network fabrics,' said Amin Vahdat, a computer science professor at UC San Diego. 'Today, the largest data centers contain over 100,000 servers. Ideally, we would like to have the flexibility to run any application on any server while minimizing the amount of required network configuration and state... We are working toward a network that administrators can think of as one massive 100,000-port switch seamlessly serving over one million virtual endpoints.'"

Comments

Re:Rehashing of long-abandoned ideas

[hairyfeet]

I think you kinda missed his point that the Networks wouldn't be so hard to admin if the corps didn't try to save a buck by lowballing and ending up with topologies that looked like they were designed by drunken gibbons. Here, let me illustrate with a true story-

So I'm working a nice little temp job, putting in a bunch of new boxes on this little insurance company when I break for lunch I run into one of my old friends at this little outdoor BBQ joint. When I tell him how easy my job is going he says "you gotta come back with me to this law firm I'm having to rebuild. You will NOT fucking believe it!" so intrigued I follow him back. On his desk are some machines, which he asked me "notice anything funny about them?" so I move the side panels so I can see and it instantly hits me that these are ALL homemade gamers rigs. He says "Yep, not a single fucking driver alike. Fun huh? And good luck with parts! But that ain't the worst part. Check this out" so he opens up the "network room" and there is literally a MOUND of Dlink and other cheap ass home routers piled up a good 4-6 feet high. I said "WTF is this?" To which he replied "This is what a dumbass who had been their "network admin" thought a network should look like. Not only is nothing labeled in this just giant fucking mess, but there are no less than SIX different ISP home plans running this shit. Fun huh?"

So while I'm sure he made out like a bandit I wouldn't have taken that job on a bet. I would have had nightmare for months trying to deal with that clusterfuck. All because some bean counter hired the first schmuck that walked through the door that could halfway talk a good game and was willing to work for the peanuts they were offering. So yeah, a network set up by someone with a brain that knows about network topologies isn't really that hard to maintain or add nodes to. But instead you get some paper tiger that can bullshit HR and makes a truly gigantic clusterfuck out of the thing and then it takes 3 forevers to get it straightened out. I don't even want to picture what kind of giant messes can be cooked up with this tech if you can just throw anything together and get it to function thanks to this "virtual mac" idea. Because when the thing finally breaks down like my buddy you might be really scared to open up that "network" door.

NATting layer two.

[argent]

They're basically NATting the layer two protocols. Combined with a super spanning tree for the natted addresses they're practically boosting layer two into layer three.

Before I read the paper I was thinking that it would be easier to just run all your services NATted at layer three, even using something like PPPoE (which is how cable networks solve the same basic problem, with something like half a million end-points on the same subnet). I guess it's more efficient to work with the simpler layer two protocols instead.