Windows Breaks Into Supercomputer Top 10

yanx0016 writes "Wow, that's some news this week at SuperComputing 08. Apparently Microsoft Windows HPC Server 2008, with a Chinese hardware OEM (Dawning), made #10 on the Top500 list, edging out #11 by only 600 Gflops. Folks were shocked to see Microsoft getting so serious around HPC; I think we are only beginning to see a glimpse of Microsoft in the HPC field."

Retarded

Honestly, why would anyone want to roll-out something like this on Windows. A lot of extra expense for little practical value.

Re:Retarded

Honestly, why would anyone want to roll-out something like this on Windows.

Did you ever see the hippos [youtube.com] doing the Dance of the Hours in Disney's Fantasia? It's like that.

Re:Retarded

Flagship demo projects like this often get exceedingly big discounts from the vendors.

Re:Retarded

I hate to defend Microsoft, but...

Crap hardware support? Who cares - you're running numerical calculations, not a bloody game on some tossy video card.

Crap vendor support? This vendor will have been given full support by Microsoft, and will be equally supportive of their users.

Performance? They're in the top 10.

Stability? If you're not dealing with odd hardware / crappy drivers, Windows Server versions are actually fairly stable.

Why not run your compute nodes under Windows?

You can actually run Windows Server 2000 and above headless, removing any GUI overhead - so why not?

I still agree that on any particular hardware configuration, Linux or another *nix will likely be faster, but your experience of desktop applications doesn't necessarily translate to HPC.

Re:there are lots of Windows developers out there.

It is possible, even if not entirely likely.

Developing a language and compiler that takes advantage of multiple CPUs (especially if it's scaling the number of CPUs) is something that a lot of research (or money). MS does have this. Whether they use it effectively is another matter.

Also, remember that they are not unfamiliar with HPC abstraction. Direct3D abstracts the architecture of the GPU, and GPUs have been parallel processors for a decade or so.

Re:there are lots of Windows developers out there.

Problem is they've missed the boat. Linux already has compilers for multiple CPUs

Look at this chart..

http://www.top500.org/stats/list/32/os [top500.org]

Windows HPC 2008 is on 4 machines out of 500. (+1 is windows 2003 if you want to count that)
Linux is on 454 out of 500 super computers

Which Operating System do you think is going to have better tools to support Super Computing?

Also I am hoping you mentioned Direct3D as to get a point across and you're not suggesting that Direct3D be used on these machines?

Re:there are lots of Windows developers out there.

Yes, actually. There are many concurrency projects for .NET. Take a look at declarative languages like F#, PLINQ (parallel LINQ), Parallel C#, Polyphonic C#

http://en.wikipedia.org/wiki/F_Sharp_programming_language [wikipedia.org]

http://en.wikipedia.org/wiki/PLINQ [wikipedia.org]

http://www.parallelcsharp.com/ [parallelcsharp.com]

http://research.microsoft.com/~nick/polyphony/ [microsoft.com]

All I read was "Windows Breaks"...

...and I thought "hey, that's not news. I've known that for years!"

Helping power the Great Firewall of China!

From the Dawning site:

Arming the "Golden Shield" project with comprehensive IT technology
With the rapid development of the Internet, the public security information construction has become an important component of national information construction. Dawning made contributions in improving information technology level within all of the public security departments, arming the "Golden Shield" project with information technology, equipping the "police" force with digitalization, intensifying the police by technology and comprehensively raising China public security's law enforcement and administrative capacity.

I like how they quote "police" force.

Cost per MIP or how many CALs

So how many CALs are required to access the system?

And if I want to make the system available to a different researcher every 2 hours how much is it going to cost them to be license compliant?

How much cpu power am I going to need to compute the licensing costs?

http://www.microsoft.com/resources/sam/lic_cal.mspx

From the article, pricing is

"With the release of HPC Server 2008 a few weeks ago, Microsoft also offered an academic version priced at $15 per node to generate interest. By comparison, a commercial license runs $450 per node"

Re:From the article, pricing is

Each node probably has 4 CPUs and 4 cores each, which reduces the price significantly, to only $28 for the commercial version, or about a dollar per node for the academic version.

That's not bad. And of course you don't understand the CALs, but hey, making erroneous statements can get you modded insightful so maybe I should spout something disingenuous about Linux, like it costs $699 to license it from SCO or something.

(For the uninformed, not all CALs are created equally and the parent is assuming that these are named licenses that must be purchased for each user. Many different kinds of CALs exist, and I suspect these are either physical unit licenses or concurrent access licenses, i.e.: you purchase 1 per node, period.)

McColo

Shortly after coming online, they noticed that it broke a speed record downloading "instructions" from abilena.podolsk-mo.ru

Off topic, but I have to mention it

edging out #11 by only 600 Gflops

Emphasis mine.

Maybe I'm suffering from a case of advancing years, but I couldn't help but be amazed by this metric. These days it is indeed small, but another part of me remembers being a fifteen year old kid amazed at how absolutely great his C64 was.

I wonder exactly how many years a C64 would have to run to make up a single seconds worth of that difference. How long would a C64 have to run to perform 600 Gflop? How long would every single C64 ever made have to run? I wonder.

You'd have to run some integer-only 6502 IEEE floating point library or something like that to figure out how long a single floating point operation would take on the C64. Then multiply by 600G.

Would it be a few years? A few millenia? Blue-green algae?

Re:Off topic, but I have to mention it

Ok, just because I'm strange I had to go and figure it out.

A C64, according to this guy [canberra.edu.au] runs at about 320 flops.

So, it would take that C64 600*10^9 / 320 = 1,875,000,000 seconds. That's 59.46 years.

Wiki says there were 30 million C64 units ever made. [wikipedia.org]

So that would be 1,875,000,000 seconds / 30,000,000 = 62.5 seconds.

It would take every single C64 ever made about a minute to make up the difference.

Wow.

Crap I'm old. =)

Re:Off topic, but I have to mention it

A C64, according to this guy runs at about 320 flops.

That just can't be. I remember the Programmer's Reference Manual showing most normal instructions finishing in 2 or 3 clocks, or maybe 350,000 IPS. I can't imagine that FLOPS would be 1,000 times slower than other opcodes. I mean, I'm pretty certain I could re-implement them in assembler in many fewer than 1,000 instructions.

Hopefully, HP will like this

For once, a computer that deserves the "Vista capable" sticker.

Potentially bogus

A couple of years ago I was surprised when one of my HPC customers issued a press release saying that their machine ran Windows HPC. The high-speed interconnect we'd sold them had no Windows drivers. You can guess what was going on: MicroSoft paid for the press release, and the machine actually ran Linux.

Dawning's previous fast machine ran Linux.

Re:Potentially bogus

What is most interesting to me is that in the case of HPC, the situation between Windows and Linux is reversed. Linux has overwhelming market share in HPC, compared to Windows status as a niche player (and that is being generous). Despite this fact, Microsoft regularly gets fawning coverage in the media for their HPC efforts, far more than they should be if you consider their marketshare. It's like PC Magazine going on and on about all the latest developments in the Linux desktop market.

Re:Potentially bogus

Perhaps if "Linux" employed the ad agency Microsoft did you'd be seeing those articles?

Windows systems are in top500 are declining

What's missing in the article is that there are only a few windows-based systems in the top500 and there numbers have been declining over the years.

Obligatory

Can you imagine a botnet of those?

I can.

Re:Yeah, mut how much useful stuff is happening?

FLOPS and MIPS are all very well, but if the OS is pissing them away then it does not matter much.

(Interviewing MS HPC Program Manager)

"Well, yeah it does stuff! Just look. You've got it all right here...Word, Excel, even Access. And just wait until you see how fast the cards fly when you win Solitaire!"

The OS is very important

If you have a huge multi-CPU multi-threading system then internal OS data structure scalability and performance are very important for anything except very trivial applications. "OS pissing" basically acts as a scaling function for Amdahl's Law.

It is one thing to measure Drystones etc, or some other simple grunt-measuring metric, but that does not realistically stress the OS's influence on how the system will perform on huge complex number crunching models.

Microsoft has only been in this game for a short time and only recently got support for 256 cores. Getting support is one thing, getting **good**, optimised, support is quite another and that will take some time to get right.

Re:i must

But does it run Linux?