why do I always think Omni Consumer Products when I hear of OCP taking over the world!
Not content with shaking up storage and servers, Facebook is creating an open source switch to help it save money on networking equipment and stop it being dependent on technologies pioneered by any single company. The switch was announced by Facebook's infrastructure czar Frank Frankovsky in a keynote speech at Interop on …
I'd buy that for a dollar!
And coincidentally, Detroit can probably be privatized pretty cheaply, and much of the city does look like a set from a certain 80s sci-fi movie franchise.
It's kind of depressing that Robocop might be the most prescient scifi movie of the modern era. When are they going to slap a laser cannon on the ISS and rechristen it as the Star Wars Peace Platform??
Dick Jones will be spinning in his grave at this misappropriation of a once great company's acronym.
"Facebook's OCP designs"
FB209: You have 20 seconds to comply... bitch!
As it's Facebook
I trust these switches will all have web-managed pages, that are just endless Status Updates, from their Friends (aka, other switches).
"Hey, that switch down on aisle A2, has been playing loose with her ports, lol"
Re: As it's Facebook
Can I haz a firmware update?
Re: As it's Facebook
Do you know this switch in a data centre on the other side of the world? [Add to network]
Available in all good shops?
So, is any of this open compute kit available to the general public or even Enterprises at all? I've not found anything available from our usual vendors; maybe I need to find an unusual vendor?
Re: Available in all good shops?
Your not generally gonna find the stuff preconfigured hardware yet or white box for openswitch. What you can find is plaes that already sell hardware thats aimed at embedded switching requirements and or bare appliances that can be adapted to ones needs.
http://soekris.com/ offers alot of embedded appliance form factor hardware for such applications. One of the nicest thins they tend to stock is a 4port gigabit ethernet card that is low power consumption, and single pcie lane. They also offer bulk discounts on everything.
There is also a nice series of tutorials here
That go over setting up a linux switch. Using open switch and vattya's routing distribution. Some skipping around is needed since he shoves alot of features in there that a baseline switch doesn't need.
You should also be able to hack some pcie extenders off ebay, to relocate the pcie lane configuration, so that you can stack 16 pcie1x cards into a chassis . Though its currently much more expensive at a build your own level. I am sure before long this will change.
Re: Available in all good shops?
I have yet to see reasonably priced hardware with multiple ports. The last time I needed something along these lines I ended up building my own.
What we need
In data centers is some forms of physical peer to peer networking connectivity... Like a ring, as it were.
Or With enough bandwidth over thin net work wires, we could drastically cut down the amount of interconnects...
There was a neighbour who used to do this sort of thing...
A bare metal switch that appears on the network like any other server...are you mad? You don't want the switch to be visible to the average rabble. I'm all for open source hardware, as long as it's cheap and still quality (a combination you don't find very often). However, some proprietary solutions are simply better than open source specifically because the hardware's designed around the software (my opinion).
It sounds like they want to create the perfect network switch that uses only industry standards (free) and open protocols. I can see it happening, even with an L3 switch and routers, but they'll have to work hard to make it robust enough where more people will want them...and I'm not sure if I'd ever want "Facebook" emblazoned across my routers -_-. I wonder who will manufacture them.
At the end of the day..
...it's a free market and you should be perfectly at liberty to buy, or not, an FB designed switch. Whether you (or I) do is in fact buy it is to some degree immaterial. What is important is the message it sends out to the big boys who have had the game field to themselves for way too long. I hope this takes off; I hope it's a success; and I welcome the increased competition. Does it really matter if your box says Facebook, Cisco or Wang Tu Industries so long as it does its job? After all, I am assuming you don't go around flashing your routers to all and sundry.
Re: At the end of the day..
No, but a lot of places are Cisco-only, and to hell with the cost to the company. And there is a huge level of job protectionism in there for 'the network guy' by requiring extensive IOS-speciific knowledge to get anything done..
Re: At the end of the day..
And when there is a problem, who is held accountable? If you are running Cisco or Juniper gear and there is a hardware or software issue, they take ownership of it. Try that with an open source product running on hardware that they didn't sell. What happens if the software side that you bought a service contract for says it is hardware, go talk to the manufacturer. You go talk to them and they say it is software, go talk to them. You have a network issue that neither side will own.
Re: At the end of the day..
That argument has been used every time a closed technology silo is under threat from open technologies and approaches, and every time its always proven false and nothing more than FUD from the under threat incumbents (and their legions of "certified" professionals).
You could equally argue that as many outages are caused by proprietary software that isn't a robust in the first place as you are entirely reliant on one vendor to code it, then support it, test multiple 1000's of configurations with every variant of other peoples technologies (let alone their own). Thats before they have to add features, design upgrade cycles etc.
What usually happens is exactly what has happened in networking, it stagnates. They play it safe and don't introduce anything revolutionary because its too much work. We still have a networking architecture designed 20 years ago .......... and it shows! It wasn't even any good back then!!
No single company can afford to build and maintain a stable platform AND drive the required innovation into it. The only way is to build an open ecosystem and divide the work up, if you architect the platform correctly in the first place then a lot of the FUD is simply irrelevant.
If you want proof of that in action just look at the x86 ecosystem, Linus or even OpenStack. All are successful and all operate on the open principle.
OCP Crime Prevention Unit 001 online.
~~ED-209 likes this.
Grow the f*** up.
Reminds me of...
Couldn't help but remember Sun's old slogan of "The Network Is The Computer" when reading this article...
Where will you find your BOFH for this
I have been a senior BOFH for years, and I wonder, when they get this new kit in, who will operate it. Well, the short answer to that is that any hacker who knows this kit will be having a heyday, and it will be years before it is nailed down enough to be truly secure. And I can't wait to google for Internet exposed open switches.
Paris - Because open kit is best.
Re: Where will you find your BOFH for this
Isn't the management problem solved with Openflow?
A web 2.0 pseudo-startup thinks it can do better than multi-billion $$$ company which has been doing this for years because ...
'Cos they buy a lot of switches, and don't like being stuck with "can only buy Cisco", but want "Can buy anything with compatible hardware and put our spin of SwitchOS on it"
- Heck, I like the idea. A couple of our suppliers make custom, industry-specific switches with industry-specific features.
Like a display on the front showing selectable details of important protocols passing through the switch, where "important" is user-definable - what matters to Facebook doesn't matter (much) to you and vice-versa.
Even in a big data centre, having a display on a switch is handy - even if merely to locate and confirm exactly which switch is having the hissy fit without relying on stuck-on labels.
Because they've gotten fat and lazy doing it? In the case of Cisco, they are overpriced, the command set changes according to the model, security updates are a pain to get your hands on, and worse yet some basic features are insecure by default and require extended options to make them secure. (SSH1 by default, SSH2 requires the Advanced encryption pack)
>... because ...
It can afford to head hunt the developers with the relevant experience and give them their heads... Remember MS and Dave Cutler ...
Yup, Cisco are a bunch of money grabbing weasels. Constantly changing the command sets, licensing models and pricing, all to make us pay more - whether it be in staff training or having to pay an extra £500 for an integrated services router cos they removed a single feature we rely on from the previous model.
IOS 15 was a joke in terms of licensing. After upgrading some of my IOS 12 routers, I lost functionality and had to go to my supplier to license it - effectively paying for a router I'd already bought!!
Its not like I can turn around and say - "ok lets rip Cisco out and bring in Juniper" - cos believe me, Id love to. Just that replacing 900 switches and routers, along with all our datacentre cores (Nexus) will be a bit of a challenge in any normal human lifetime.
........... the multi-billion $$$ company have repeatedly failed to deliver what the market wants ............ all they have done is hold the technology industry back with a legacy monolithic architecture that is not fit for purpose for the modern workloads these people run, build a monopoly that allows them to charge extortionate prices for no real reason and have not innovated for years and more importantly wont allow anyone else to because they wont let other peoples kit talk to theirs (proprietary protocol anyone!?)
Was that the answer ................ do I win a prize?
But Cisco are not the only one out there right. There are dozens on _specialist_ companies who build switches ...
Sure - being stuck on one vendor is a pain in the arse - but there are plenty of other alternatives. I'm just not sure why Facebook suddenly think they can do better. If they were doing "something special just for our workloads" then fine, but that's not what they are saying.
It just smells like not-invented-here syndrome, and if I were a shareholder I'd ask if what they are doing is really part of their core business.
Sounds like BOFH FUD to me.
The "aim is to produce a completely open sourced hardware switch, as well as being OS agnostic, where we can get full access to the firmware, and fill it full of backdoors, known only to facebook, so that we can snoop the entire planet's data." Frankovsky told us. With the switch, the OCP hopes to "break the current appliance model. Most switches today have a tightly integrated software solutions which are a total bitch to add back-doors to."
There. Fixed it for you. No charge.
But...but...i thought closed source was the one true evil riddled with back doors, and open source was the future of interoperable, peer reviewed goodness.
I'm getting too old to keep up with this game
Sounds good in theory, but how do you build true resilient networks if there's only one network. (OK, so I know Cisco, Brocade, Juniper etc won't go out of business any day soon).
But seriously, if these guys want to build enterprise or telco class switching then they need to have two completely separate Chinese Walled streams developing the same product. This greatly reduces the risk of faults and flaws being introduced - no point in having two network connections if they both have the same fault and both fail in the same conditions.
Think I'm being excessive? Look at the specifications for Linx
MASSIVE PATENT BATTLE WITH CISCO KLAXXON
Arista is already doing it
All Arista switches are basically a server with a very fast NIC card with 52 or 64 ports
You can even run a VM on the switch and use it as a boot server for the servers in the rack etc.
Its all open source and you can program / hack in python.
Re: Arista is already doing it
Replace "server with a multi-port NIC" with a custom Broadcom switching ASIC that does all the heavy lifting and a suitable (at present x86) CPU running some form of Linux for management functions and you have almost all of the 10/40/100GbE switches on the market covered with only a few exceptions (Cisco have their own custom ASIC's but increasingly rely on other vendor ASIC's to keep up with the market, Arista use Fulcrum).
This means the differences between data centre switches largely come down to product release dates and the software features offered.
Go to consumer and low-end corporate switches and you have the same with cheaper ASIC's and a lower end management processor.
If OpenFlow levels the software field for data centre switches, there will be a lot of casualties...
"And coincidentally, Detroit can probably be privatized pretty cheaply, and much of the city does look like a set from a certain 80s sci-fi movie franchise."
No it doesn't. When I was there about 5 years ago, Detroit looked far worse. After going to Detroit, the movies looked like pretty nice buildings with some movie dirt thrown on; the actual Detroit had destroyed roads, block after block where the abandoned buildings had collapsed and turned back into fields of grass, broken water mains, still-standing abandoned buildings with no windows. I expected to see gangbangers and so on but they have even left portions of the city. I went to visit a friend, and the nearest onramp to his house was a pile of rubble with "road closed" sign at the end -- it had collapsed, and the rubble was just left to sit. The next onramp had holes through it. When I drove on the highway (which is all overpass, a.k.a. elevated roadway), I could look out the window and see *through* the bridge, the tires were in fact in some cases not running on concrete at all but on the metal rebar that is supposed to be embedded under the concrete. The road I was on was so rough I hit my head on the roof a few times, and this was a 1985 Celebrity, not a car with "tight European suspension.". I would not have even considered driving a semi (a.k.a. a lorry) on these roads, I'd guess it would have been too likely to fall through! I know three other people that have been there -- one hit road debris and destroyed half of their front end, $1500. The second had an Acura with low-profile tires and cracked all 4 rims, bought replacements in Detroit, and had to replace *those* when he got back. The third came much more recently and said the roads had been fixed -- apparently some of the stimulus money congress didn't give to the banks was spent specifically on Detroit's roads (although not on the rest of the city.)
As for this switch -- sounds like a good idea, but I do bet nothing comes of it. Nothing wrong with open source development, but for hardware like this the fabrication technology for one-off builds is probably at least 10 years behind what a proudction run can use, and it's much more difficult to develop and debug something where you never even get to test it (I doubt there'll be much test silicon out and about.) There's software to run on a commodity PC and do switching and routing, but I do doubt this is what Facebook has in mind... making a "cheap and cheerful" 1gb router is just not even an issue, facebook could fix bugs in pfsense or something like this and call it a day. But once you get past that to 10gb and up gear you start needing custom ASICs and so on to maintain line speed.
Re:Requiring custom ASIC's
Google "10GbE Merchant Silicon" - most vendors are already using off-the-shelf ASIC's to avoid the time-to-market delays of creating your own.
Regarding the software - these switches are designed to provide the large scale data plane for your OpenFlow/SDN to manage. It won't be as fully featured as a comparative enterprise switch, but it probably has all the features that a typical data centre actually uses....