Look at something like a Sun T5540 if you want power/performance ratio's as well as something that is a monster of a database server and webserver. I mean, its hard to beat 256 simultaneous process threads when talking about web hosting or database hosting.
by Anonymous (not verified) on 6/25/09 at 10:34 pm |reply
sparc lost. nothing to see here. move along.
another diehard sparc fanboy yearning for the 90s good old days. the only non-x86 processor that anybody cares about is ARM, and not for much longer.
by Anonymous (not verified) on 6/26/09 at 1:29 am |reply
Why are you even thinking about Sun? He wants cheap and fast
Sun is assuredly not inexpensive, and the only thing fast about sparc systems is how quickly they drain your wallet.
Power efficient servers are easy to build. You simply give up lots of things you would otherwise get in a 1U or blade. I am guessing that the facebook person must not know how to build servers, or what goes into servers. Or he would understand that power efficient *MEANS* slow.
If this guy really wants to design large fast clusters of power efficient machines at huge scale, he should be talking to the custom HPC solution shops, that have been doing this for their customers for years. Point this guy to here and here. Ignoring some of the large companies, and focus upon those that will design the systems for you correctly from the outset.
by Anonymous (not verified) on 6/26/09 at 7:24 am |reply
flawed argument
1) Facebook & Amazon need cheap, power efficient systems
2) Intel and AMD aren't measuring up with processors to power these systems
3) However, Google has systems appropriate for this use (presumably using Intel or AMD processors)
If that's his argument, then it would seem that the real conclusion is that Facebook can't build systems as good as Google's, even though they are using the same processor technology.
by Anonymous (not verified) on 6/25/09 at 10:56 pm |reply
I understand that argument,
I understand that argument, but the truth is that this is a guy from Facebook. First, he's not speaking for Amazon so they shouldn't be in the argument. They very well may have more optimized server hardware. Second, Google is a company making many BILLIONS of dollars worth of profit. Facebook has yet to make one. Google can afford to design (or pay for someone to design) their own hardware platforms. Facebook doesn't have this option. What he was basically saying is, if Google can do it, why can't an OEM offer this for the many other companies who don't have that money? Google uses commodity hardware to reach this goal so it should be relatively easy and profitable for OEM vendors to do this also.
by Anonymous (not verified) on 6/26/09 at 12:11 am |reply
Missing the point
Or perhaps Google a seemingly larger organization can better exploit economies of scale so they they can afford to do their own R&D and even some fab while facebook has to rely upon the offerings of HP & Sun etc.
by Anonymous (not verified) on 6/26/09 at 12:13 am |reply
Facebook doesn't know how to mange servers
That seems to be the case and I agree completely. So basically, facebook is going crazy try to find a solution to the $300 million dollars of servers that they are using. So they try buying new Intel and AMD processors, only to find out they arn't as good as advertised...duh
by Anonymous (not verified) on 6/26/09 at 1:00 am |reply
It's the whole system...
It's not just about the processor, it's about the entire integrated system. From the power supply, to the motherboard layout, to what voltages are actually necessary to send to the motherboard, ...
Google designs and builds their own power supplies, motherboards, etc., to be power efficient (http://news.cnet.com/8301-1001_3-10209580-92.html). No OEM comes close to doing anything like it.
by Anonymous (not verified) on 6/26/09 at 1:47 am |reply
RE:flawed argument
However, Google has systems appropriate for this use (presumably using Intel or AMD processors)
If I had the cash, I'd roll out dual core 1.6Ghz Atom boards. They have a small form factor, and I bet you could cram loads of them into a custom rack ala Google-style. Some might argue that the Atom's don't have enough juice, but if you consider the power consumption, the form factor I think it's a viable solution.
by Anonymous (not verified) on 6/26/09 at 4:59 am |reply
Google uses a single input voltage
Google's custom motherboards take a single 12v input voltage then use high efficiency onboard transformers to provide the various voltages required by different subsystems.
Having the power supply provide all those voltages (like it's done in servers from HP, Dell, etc.) is much less efficient. So it's not a flawed argument based on the commonality of Intel and AMD processors. A lot of power is lost in your 80% efficient AC Power supply and making the power supply more efficient isn't the best solution.
But, yes, it's true, most companies don't have and can't justify the engineering resources to custom design system boards like Google has.
by Anonymous (not verified) on 6/30/09 at 8:17 pm |reply
Bad web application code and
Bad web application code and design == Bad performance on any web server.
by Anonymous (not verified) on 6/26/09 at 12:03 am |reply
New features
"Testing is another key to the operational success of Facebook, which has more than 200 million users and frequently introduces new features. Heiliger said the launch of Facebook's personalized usernames earlier this month went smoothly, despite an explosive response when it first went live, because of extensive testing of the new feature."
Yes, congratulations! You gave people a username. That's a fantastic achievement - and a very useful one as well. Great to have such forward thinking in the world.
by Anonymous (not verified) on 6/26/09 at 1:38 am |reply
Dear Facebook
Could It Bee
Your Crummy
Pee Haitch Pee?
by kosh (not verified) on 6/26/09 at 4:35 am |reply
Cheap?
He says they need lots of servers so "the servers have to be cheap," But really they have to buy servers regardless of the price if they want to keep up with the growth so no need for server makes to make them cheap.
Notice later he also says "There's a pretty simple answer for scaling infrastructure. It's, 'Don't be cheap,'"
quality doesn't come cheap.
by the way, it sucks that after I preview if I make another change I have to re-enter the captcha. And why can't I get blank lines between paragraphs?
by Edward (not verified) on 6/26/09 at 9:11 am |reply
Double captcha
Edward, thank you for the feedback on the capture re-enter and line spacing. Changes are coming soon!
So what he's saying is that Google builds themselves some great data centers using existing Intel and AMD processors, but Facebook cannot do the same? Seems like his criticism would be better directed at the designers of the server platforms they buy and / or the designers of their data centers.
They're trying to find "in real time" why they're not getting the perf increases they expected. I'm guessing it's unreasonable expectations, combined with sub-optimal application architecture. The apps are probably not as able to efficiently take advantage of all the cores in a system as they thought they were, and that has more to do with software architecture than processor microarchitecture. The processor microarchitectures have been well known and documented for years before they were released, and my bet is they haven't properly designed apps to take advantage of their power.
by Hick Ninja (not verified) on 6/26/09 at 9:44 am |reply
Maybe it's the operating system
The article doesn't specify which operating system they are working with.
The chickens are starting to come home to roost for Windows and Linux. Now that chips from Intel and AMD are capable of supporting dozens of threads in a single system, it follows that the operating system must scale to high thread count.
Linux and Windows are inadequate in this area. Solaris, on the other hand, scales to hundreds of threads due to Sun's experience in scalability on very large SMP SPARC platforms, and Sun is now reaping the benefit of that investment on 4-socket Intel/AMD systems with 6 cores per socket.
The "SPARC fanboy" in an earlier post may have missed the mark on hardware, but he's right on target with the OS choice.
by John Slater (not verified) on 6/26/09 at 5:06 pm |reply
Sidekick: The Good News & the Bad News Either way you look at it Microsoft Data Center management did not follow standards or best practices in this failure. In which case it makes me wonder more about the outsourcing of corporate data much less personal data.
- mburton325
Surviving Windows is easier than you think… MKS offers the power of an integrated all-in-one environment and provides you with the Power of UNIX on Windows Learn More
Brought to you by:
Free books
We have 5 copies of these two new books to give to some lucky readers. The deadline for entries is November 30, 2009.
AISO founders envisioned a Web hosting company that was environmentally friendly. While the company employed energy-efficient innovations like solar panels, its infrastructure produced unacceptable power and cooling requirements. Find out how AISO leveraged AMD technology to overcome their challenge in this case study white paper.
In this whitepaper, Scalar explores the opportunity to change the landscape with respect to mission critical databases
built around Oracle. Leveraging technologies such as Linux, high-end commodity processing power and Oracle RAC
technology to architect, design, build and maintain database infrastructure that delivers maximum availability, reliability
and performance at a fraction of traditional cost.
On a typical day, weather.com, the Web site for The Weather Channel in Atlanta, serves up between 15 million and 20 million page views. But in September 2004, when back-to-back hurricanes ransacked Florida, the peak traffic on one day more than tripled: over 70 million page views by more than 7 million unique visitors. Read the full success story now.
jfruh
pasmith
Esther Schindler
mikelgan
David Strom
sjvn
Sandra Henry-Stocker
Quick, practical advice for IT pros. Made fresh daily.
Someone is looking at the wrong technology....
Look at something like a Sun T5540 if you want power/performance ratio's as well as something that is a monster of a database server and webserver. I mean, its hard to beat 256 simultaneous process threads when talking about web hosting or database hosting.sparc lost. nothing to see here. move along.
another diehard sparc fanboy yearning for the 90s good old days. the only non-x86 processor that anybody cares about is ARM, and not for much longer.Why are you even thinking about Sun? He wants cheap and fast
Sun is assuredly not inexpensive, and the only thing fast about sparc systems is how quickly they drain your wallet.Power efficient servers are easy to build. You simply give up lots of things you would otherwise get in a 1U or blade. I am guessing that the facebook person must not know how to build servers, or what goes into servers. Or he would understand that power efficient *MEANS* slow.
If this guy really wants to design large fast clusters of power efficient machines at huge scale, he should be talking to the custom HPC solution shops, that have been doing this for their customers for years. Point this guy to here and here. Ignoring some of the large companies, and focus upon those that will design the systems for you correctly from the outset.
flawed argument
1) Facebook & Amazon need cheap, power efficient systems2) Intel and AMD aren't measuring up with processors to power these systems
3) However, Google has systems appropriate for this use (presumably using Intel or AMD processors)
If that's his argument, then it would seem that the real conclusion is that Facebook can't build systems as good as Google's, even though they are using the same processor technology.
I understand that argument,
I understand that argument, but the truth is that this is a guy from Facebook. First, he's not speaking for Amazon so they shouldn't be in the argument. They very well may have more optimized server hardware. Second, Google is a company making many BILLIONS of dollars worth of profit. Facebook has yet to make one. Google can afford to design (or pay for someone to design) their own hardware platforms. Facebook doesn't have this option. What he was basically saying is, if Google can do it, why can't an OEM offer this for the many other companies who don't have that money? Google uses commodity hardware to reach this goal so it should be relatively easy and profitable for OEM vendors to do this also.Missing the point
Or perhaps Google a seemingly larger organization can better exploit economies of scale so they they can afford to do their own R&D and even some fab while facebook has to rely upon the offerings of HP & Sun etc.Facebook doesn't know how to mange servers
That seems to be the case and I agree completely. So basically, facebook is going crazy try to find a solution to the $300 million dollars of servers that they are using. So they try buying new Intel and AMD processors, only to find out they arn't as good as advertised...duhIt's the whole system...
It's not just about the processor, it's about the entire integrated system. From the power supply, to the motherboard layout, to what voltages are actually necessary to send to the motherboard, ...Google designs and builds their own power supplies, motherboards, etc., to be power efficient (http://news.cnet.com/8301-1001_3-10209580-92.html). No OEM comes close to doing anything like it.
RE:flawed argument
However, Google has systems appropriate for this use (presumably using Intel or AMD processors)The argument was directed at server vendors.
Intel Atom
If I had the cash, I'd roll out dual core 1.6Ghz Atom boards. They have a small form factor, and I bet you could cram loads of them into a custom rack ala Google-style. Some might argue that the Atom's don't have enough juice, but if you consider the power consumption, the form factor I think it's a viable solution.Google uses a single input voltage
Google's custom motherboards take a single 12v input voltage then use high efficiency onboard transformers to provide the various voltages required by different subsystems.Having the power supply provide all those voltages (like it's done in servers from HP, Dell, etc.) is much less efficient. So it's not a flawed argument based on the commonality of Intel and AMD processors. A lot of power is lost in your 80% efficient AC Power supply and making the power supply more efficient isn't the best solution.
But, yes, it's true, most companies don't have and can't justify the engineering resources to custom design system boards like Google has.
Bad web application code and
Bad web application code and design == Bad performance on any web server.New features
"Testing is another key to the operational success of Facebook, which has more than 200 million users and frequently introduces new features. Heiliger said the launch of Facebook's personalized usernames earlier this month went smoothly, despite an explosive response when it first went live, because of extensive testing of the new feature."Yes, congratulations! You gave people a username. That's a fantastic achievement - and a very useful one as well. Great to have such forward thinking in the world.
Dear Facebook
Could It BeeYour Crummy
Pee Haitch Pee?
Cheap?
He says they need lots of servers so "the servers have to be cheap," But really they have to buy servers regardless of the price if they want to keep up with the growth so no need for server makes to make them cheap.Notice later he also says "There's a pretty simple answer for scaling infrastructure. It's, 'Don't be cheap,'"
quality doesn't come cheap.
by the way, it sucks that after I preview if I make another change I have to re-enter the captcha. And why can't I get blank lines between paragraphs?
Double captcha
Edward, thank you for the feedback on the capture re-enter and line spacing. Changes are coming soon!misdirected criticism
So what he's saying is that Google builds themselves some great data centers using existing Intel and AMD processors, but Facebook cannot do the same? Seems like his criticism would be better directed at the designers of the server platforms they buy and / or the designers of their data centers.They're trying to find "in real time" why they're not getting the perf increases they expected. I'm guessing it's unreasonable expectations, combined with sub-optimal application architecture. The apps are probably not as able to efficiently take advantage of all the cores in a system as they thought they were, and that has more to do with software architecture than processor microarchitecture. The processor microarchitectures have been well known and documented for years before they were released, and my bet is they haven't properly designed apps to take advantage of their power.
Maybe it's the operating system
The article doesn't specify which operating system they are working with.The chickens are starting to come home to roost for Windows and Linux. Now that chips from Intel and AMD are capable of supporting dozens of threads in a single system, it follows that the operating system must scale to high thread count.
Linux and Windows are inadequate in this area. Solaris, on the other hand, scales to hundreds of threads due to Sun's experience in scalability on very large SMP SPARC platforms, and Sun is now reaping the benefit of that investment on 4-socket Intel/AMD systems with 6 cores per socket.
The "SPARC fanboy" in an earlier post may have missed the mark on hardware, but he's right on target with the OS choice.