Sign in with
Sign up | Sign in

Facebook VP Slams Intel, AMD Server Chips

By - Source: Tom's Hardware US | B 45 comments

Facebook's Vice President of Technical Operations says that new chips from AMD and Intel don't deliver promised performance increases.

During an interview with GigaOm founder Om Malik at GigaOm's Structure '09 conference, Facebook VP of Technical Operations Jonathan Heiliger was asked about unexpected problems the company had experienced while trying to keep up with Facebook's growing population.

"The biggest thing (that) surprised us is ... less-than-anticipated performance gains from new micro-architectures -- so, new CPUs from guys like Intel and AMD. The performance gains they're touting in the press, we're not seeing in our applications," Heiliger told Malik. "And we're, literally in real time right now, trying to figure out why that is."

Heliger went on to declare that OEMs just "don't get it."

"You guys don't get it. To build servers for companies like Facebook, and Amazon, and other people who are operating fairly homogeneous applications, the servers have to be cheap, and they have to be super power-efficient," said Heiliger. "And that doesn’t just mean putting in a really highly efficient power supply. It means going all the way down, basically starting at the wall outlet, all the way to the processor and figuring out how to optimize that power path,” he continued. "Google has done a great job designing and building its own servers for this kind of use."

Check out the full video of the interview here.

Display 45 Comments.
This thread is closed for comments
Top Comments
  • 28 Hide
    Hanin33 , June 26, 2009 9:46 PM
    that's the price these guys pay for trying to use comoddity parts in place of mainframes and true iron... do they really expect large scale energy efficiency with a surplus of computational power from a system that could easily be bought by a consumer? who's really the deluded one here?
  • 28 Hide
    mcnuggetofdeath , June 26, 2009 10:02 PM
    Facebook has had issues with outages since before Nehalem launched. The guy is just trying to cover his own ass, and thats not cool. They shouldve dumped more money into infrastructure, rather than redesigning something that already works or lining your own pockets with the craploads of ad revenue they generate. And to use someone elses assesment of how a processor performs within a given situation, and expect it to scale indefinitely is a sign of his incompetency. I hope he gets fired.
  • 21 Hide
    Gutbop , June 26, 2009 11:07 PM
    The guy talks about how AMD and Intel processors aren't giving him the performance he wants, but then talks about how Google designs and builds servers that are efficient??

    Did I miss something? Google doesn't build their own processors. They still use Intel or AMD (or IBM) processors. They just design and put together the server.

    Maybe what he really means to complain about is that his people don't design their servers well enough, and he wants Google's people to design his servers instead.
Other Comments
  • 10 Hide
    jhansonxi , June 26, 2009 9:45 PM
    Maybe Google should get into the server business. I could see their name on the side of shipping containers full of super-efficient servers.
  • 28 Hide
    Hanin33 , June 26, 2009 9:46 PM
    that's the price these guys pay for trying to use comoddity parts in place of mainframes and true iron... do they really expect large scale energy efficiency with a surplus of computational power from a system that could easily be bought by a consumer? who's really the deluded one here?
  • 28 Hide
    mcnuggetofdeath , June 26, 2009 10:02 PM
    Facebook has had issues with outages since before Nehalem launched. The guy is just trying to cover his own ass, and thats not cool. They shouldve dumped more money into infrastructure, rather than redesigning something that already works or lining your own pockets with the craploads of ad revenue they generate. And to use someone elses assesment of how a processor performs within a given situation, and expect it to scale indefinitely is a sign of his incompetency. I hope he gets fired.
  • 4 Hide
    curnel_D , June 26, 2009 10:18 PM
    "lining your own pockets with the craploads of ad revenue they generate"

    I could have sworn that's what they're in the buisness to do.

    Facebook doesnt owe us anything.
  • 20 Hide
    paranoidmage , June 26, 2009 10:47 PM
    I think an executive like him should know better than anyone that AMD and Intel are trying to sell a product and only showing benchmarks that make them look good. Not only that, but they are only benchmarks and not true indicators of performance. If he is truly as stupid as he makes himself out to be, Facebook should think about replacing him.

    "The performance gains they're touting in the press, we're not seeing in our applications"

    Maybe he should try optimizing his applications for the microarcitecture instead of expecting the microarchitecture be designed to fit his needs perfectly.
  • 5 Hide
    ph3412b07 , June 26, 2009 10:53 PM
    has anybody tried overclocking a blade server?..... lol
  • 9 Hide
    caqde , June 26, 2009 10:57 PM
    Quote:
    Maybe he should try optimizing his applications for the microarcitecture instead of expecting the microarchitecture be designed to fit his needs perfectly.


    Yeah that would likely be the issue. Todays server processors have many cores, if you want your server application to get a speed boost it is time you talk to your engineers working on facebook to start optimizing facebook for many processors. Of course they need to make sure any API's and server software they use is multi-core friendly too. Maybe they can start laying the blame somewhere else.
  • 21 Hide
    Gutbop , June 26, 2009 11:07 PM
    The guy talks about how AMD and Intel processors aren't giving him the performance he wants, but then talks about how Google designs and builds servers that are efficient??

    Did I miss something? Google doesn't build their own processors. They still use Intel or AMD (or IBM) processors. They just design and put together the server.

    Maybe what he really means to complain about is that his people don't design their servers well enough, and he wants Google's people to design his servers instead.
  • 5 Hide
    Lans , June 26, 2009 11:43 PM
    Maybe they should make their own CPUs for their
    Quote:
    fairly homogeneous applications
    like
    Quote:
    “Google has done a great job designing and building its own servers for this kind of use”
    .

  • 20 Hide
    battery , June 26, 2009 11:55 PM
    Quote:
    Heliger went on to declare that OEMs just “don’t get it.”

    clearly they should create a new line of processors specifically for facebook.

    alternatively, stfu and optimize.
  • 1 Hide
    rbarone69 , June 27, 2009 12:18 AM
    They should really find some good engineers to figure out where their bottlenecks are. I'll bet the system has so much scope creep they have no clue where the performance problems are and this VP is just spewing crap some ignorant director / admin / programmer gave him to excuse the problems.

    The new Nehalem Xeon's are configured for dual processor sytsems and are 4 cores at the moment (I think). Soon there will be 16 thread executions per processor (8 cores, 8 HT). If programmers know how to use them they are VERY VERY fast. With cutting edge systems you make due with what you have and plopping core i7's in server environments could be a bad choice.

    I'm sure this guy is simply ignorant (not his fault, I blame his staff and hiring). Go buy some low power xeons, restructure slow parts of your code, and stop whining!

  • -5 Hide
    Anonymous , June 27, 2009 12:57 AM
    Could their developers be to blame? SRSLY, AMD and Intel have both been delivering the hotness lately... Hyperthreading is a complete gimmick and utter shite, but Nehalem and Shanghai are both still quite robust...
  • -4 Hide
    Anonymous , June 27, 2009 2:06 AM
    They better go with an IBM server if they want to keep up with updating statuses and applications for every person on the planet!
  • 1 Hide
    Vettedude , June 27, 2009 2:14 AM
    Maybe Facebook should defrag their drives every once in a while. That's where the bottleneck in their server would be.
  • 0 Hide
    zodiacfml , June 27, 2009 3:52 AM
    he's just blaming the hardware guys. google did not expect from those anyway and i wonder what CPUs they're using. :) 
  • 2 Hide
    nachowarrior , June 27, 2009 5:19 AM
    First of all to be power efficient the software running on the processors needs to be optimized. If software these days wasn't so bloated we could run a lot more on a single server. Secondly, if configured properly you can get quite the bang for your buck (power and money). I think there's more likely an issue on Facebooks part rather than on the hardware level. He's just ticked off because google patented the idea to put ups' on a per/board basis.
  • 1 Hide
    the last resort , June 27, 2009 5:43 AM
    they need to reoptimize their code to take advantage of the processors. Sorry, but a single threaded game will be improved going to a quad core. The game will still only use 1 CORE. You got that VP?
  • 1 Hide
    battery , June 27, 2009 6:07 AM
    Quote:
    "The performance gains they're touting in the press, we're not seeing in our applications"

    I'd also like to point out when a company, for example, says something is up to 50% more efficient, they mean it's up to 50% more efficient.
    Just because facebook code written by a highschool grad doesnt fully utilize an architecture doesnt mean the performance isnt there.
  • -7 Hide
    buzznut , June 27, 2009 11:25 AM
    Yeah, because facebook is more demanding than crysis or any other application, it keeps dropping frames because social networking is SO computing intensive. Does anyone remember the old Hustler magazine a@$hole of the month? Not that I've ever read that filth...

    Someone needs to explain to this jerkoff that facebook is on its way out, its 15 minutes is up.
Display more comments