Slashdot Mirror
Microsoft Competes In Supercomputer Market
HoboMaster writes "Microsoft is releasing a public beta of Windows Compute Cluster Server 2003 in their first attempt to compete in the supercomputer OS market. Gates is planned to speak at the 2005 Supercomputer Conference, which will be Microsoft's first appearance at the conference. Gates, as always, has high hopes for this new version of Windows, even claiming it to be as powerful and easier to use than Linux."
Supercomputers aren't about "Ease of use." They're about speed per dollar. When WCC can beat Linux on price/performance, then people will stand up and take notice. Not before.
"He who would learn astronomy, and other recondite arts, let him go elsewhere. " -- John Calvin, commenting on Genesis 1
I don't know what the "easy to use" argument precisely means in this arena. The people assembling and managing large clusters to produce supercomputers are in a significantly different league than your average MS-Word user. Not to poo-poo MS or anything, but Gates seems to have forgotten the audience in this case.
The world's burning. Moped Jesus spotted on I50. Details at 11.
In what way? It seems the two are equivalent- double click this icon to start a program. Menu bar over there for a list of all programs. Except Linux has a good text interface as well as the GUI. If anything, that gives the edge to Linux.
I still have more fans than freaks. WTF is wrong with you people?
It's hard to see how MS will gain much market penetration in this area unless it substantially alters its licensing. As I said in another post, "ease of use" isn't exactly going to be a big selling point for the guys putting these kinds of computers together, and OS licensing costs of $0.00 is hard to beat.
The world's burning. Moped Jesus spotted on I50. Details at 11.
he's actually claiming it's *as powerful* as linux!? why should that interest anyone, unless it's also as free?
Nothing says "Product of the Future" like the beta version of a product named "Windows Compute Cluster Server 2003".
Sure, they're doing it to maintain the "2003" branding of the flagship server. But why, less than two months before the end of 2005, are they not even trying to sound modern?
I have to say, I just don't think developers will go for it. First of all, if you're setting up a cluster of, say, 20 machines to run some MPI programs, you're going to be funneling some serious coin microsoft's way.
Secondly, I, like many developers, have been running MPI programs on Linux clusters for some time now. What's my incentive to switch? All I've got is penalties, like having to buy software and stuff. MPI is already free, open source software. So now MS sticks it in their OS and sell it as a new platform?
At least for me, this is too little, too late. I'll do what I've been doing, which is run my parallel code on Linux.
bottom line is, until microsoft can build this OS to be HUGELY FASTER than linux, there's no reason to pay extra for something that doesn't have any speed advantages.
i've never heard of the supercomputing crowd complaining about ease of use, they are looking for more calculations for less money, and for that linux/unix is probably still the best choice. there's no reason to pay thousands for an OS that doesn't increase your performance any further than an OS that costs $0
When I worked at Microsoft, I specifically argued *against* getting involved in HPC markets. It isn't really an interesting market. It isn't a big market. And it never will be. If anything it will get smaller rather than bigger. Yes, there are some applications that are not going away but these are not common. After all how many customers does Cray have? How many customers does Microsoft have? Ok, you have the answer to my question. Heck, the ISP and web presence provider markets are more important to Microsoft strategically than HPC.
Indeed I cannot think of *any* reason why one would want Windows on an HPC cluster. Indeed, with Microsoft's reliance on COM and IPC stuff, I would be highly skeptical of using the Windows development environment in these cases. Yes, async I/O might be more mature on Windows, but I think that on the whole, Linux is a better choice.
As for the ease of use factor. This is a product that is really only needed by a few highly technical people. Ease of use for beginners is not important here. Ease of use by experienced UNIX admins is. Sadly Windows fails here pretty badly. After all not everyone needs to build a Beowulf cluster with licensed Windows software in their basement and the intensive number crunching apps that such clusters are used for are the exception rather than the rule.
Finally.....
Why not take Windows Server 2003 Standard or even XP Pro (for fewer than 10 nodes), install SFU 3.5 and PVM and build your cluster that way? It seems that this would be better for the market than this new product which seems to be the worst of both worlds.
This is just about saying "Anything Linux can do Windows can do better" rather than pursuing any reasonable business plan.
LedgerSMB: Open source Accounting/ERP
1. PHB sees Microsoft adverspamming for Windows Computer Cluster 2003, and believes the drivel. ...
2. PHB makes case to execs, gets capital for an 80-node WCC2K3 cluster for eleventy billion dollars, thanks to Licensing 7.
3. Admins shake their heads in disdain, get the thing running, and walk away.
4. Developers waste time and resources reinventing the wheel.
5. Nodes start to get rooted because the admins didn't harden the system.
6. Organized crime groups use nodes to DDoS websites in the name of extortion.
7.
8. Profit! (for Microsoft, at least).
"Ancillary does not mean you get to rule the world." --U.S. Circuit Judge Harry Edwards, speaking to the FCC's lawyer
They bring very little value to the back end. A typical distributed app is not a rich UI client that needs lots of Windows APIs to play DRM'd movies, so Windows has no advantage there. It's a C, C++, or Fortan (mixed, even) job running MPI over some specialized interconnect hardware.
You also need a good parallel file system which I'm willing to go out on a limb and say that CIFS is probably not the optimal choice - any real system will probably be using a dedicated filer.
However, their strong suit in this space are tools like IDEs. If they can convince folks that using Windows as a front-end to development, then they can make some good inroads.
Right now the supercomputing folks are starting to get interested in Eclipse, and they're trying to head that off, not to mention small ISV's like us.
I can explanate how to administrate your network. You must configurate and segmentate it, so it can computate.
And even if MS were so compelled, what really is in it for Redmond? It's probably the smallest market in the world, with a customer base measured in the thousands, and one that already has access to either operating systems with a long multicomputing heritage or to clustered Linux systems that, for the kinds of guys that set up and maintain supercomputer clusters, offers no unreasonable difficulties as far as usability.
I think what we have here is Gates' inferiority complex towards Linux. He desperately wants to have MS in every market, and can't stand the thought of that open source demon increasingly being utilized, so he'll waste money on a product that, even if it were to be a success within that marketplace, would provide an outrageously small amount of revenue.
I suppose MS has money to burn, but if I had that much money to piss away, I'd try to take over the pocket calculator or security alarm markets.
The world's burning. Moped Jesus spotted on I50. Details at 11.
A standard line from Bill, "wait till you see Vista its better", "wait till you see compute its faster".
Amazingly the press continue to take Microsoft at face value on annoucing their version as better when they don't release what they announce.
So sure MS is better at supercomputers... I mean they have such a history in it, just look at the top 500 its just littered with MS boxes.
This isn't Windows v Linux, this is MS Research v IBM Research. The people behind the CPU, Relational databases, reliable messaging and of course the huge amount of work on massively scalable computers. If MS had real ability they'd be working with the big processing boys from the goverment and weather prediction areas.
An Eye for an Eye will make the whole world blind - Gandhi
Security issues are irrelevant in a lot of cases. Scientific computing isn't done on computers attached to the internet...it is done on intranets consisting of specialized hardware streamlined for the needs of HPC. Most HPC programs don't even attach themselves to ethernet networks, but rather to things like Myrinet (bypass OS calls to reduce overhead GREATLY) that are intended for HPC. Being DDoS'd, or having a 'zombie cluster' etc are not really issues here.
I think the advantage of a MS solution might be ease-of-use, especially in server clusters that are up for hire (that is, up for timesharing). If you are some group performing research that requries lots of power but aren't focused in a CS-related field, you may not have the resources to go use the (often arcane) parallel (MPI) debuggers etc. and churn out a top-grade program for a supercomputer. An MS solution might indeed be cheaper OVERALL because of time-to-solution (time = money). Let's face it, VS.NET is a dream to code in - compared to other well-featured IDE's like Eclipse, it is light-weight, easier to use (Eclipse has major bloat issues), etc. So who knows - as the article mentions, it might indeed become part of an end-to-end scientific process, where the computational parts seamlessly fit in.
Furthermore, everyone who is talking about licenses per processor are not thinking properly...do you really think they would achieve penetration with the barrier to using the software so high? Of course not! Instead of speculating negatively, let's just wait and see what the licensing programs are when the product is released.
My 2 cents
- Prestige (branding)
- Research
For whatever Microsoft spends improving Windows so that it can be used on today's supercomputers, the benefits they will reap for their server and workstation lines could easily repay that investment. I'm not at all sure they can make a go of it, but if they succeed it will help them a lot more than just selling some licenses for big iron.Regardless of what the folks who actually use the supercomputer want, there's always this administrator who signs off on the purchase who will say 'Windows, huh? Great. Now we can have one support contract that covers everything!'. The M$ Sales Rep takes him/her out to a couple fancy lunches and comes back with a signed contract.