Microsoft testing Excel for supercomputers

At a key supercomputing conference on Monday, Microsoft released a test version of its Excel spreadsheet redesigned to run on powerful clusters of servers.

By engineering Excel to run better on such clusters Microsoft said that customers are seeing spreadsheets that normally would take weeks to calculate now run in a few hours.

The software maker also released a beta version of Windows HPC Server 2008 R2--the latest version of Windows Server designed to run in high-performance compute clusters. The announcements were made at the SC09 conference in Portland, Ore.

Microsoft has taken the standard version of Excel 2010 and combined it with new Windows HPC Server 2008 R2 technology, allowing Excel to run on the cluster. The final version of Excel compute cluster and Win HPC Server 2008 R2 is expected to be ready in summer 2010. The capability has been in development for about 18 months.

The announcements are the latest in Microsoft's push over the last few years to better compete against Linux in the market for compute clusters--high-performance systems built by linking together large numbers of standard servers. Last year, for example, Microsoft managed to crack the upper echelons of the supercomputing ranks, landing in the top 25 rankings for the first time.

Microsoft also said the next version of its developer tools--Visual Studio 2010--will help ease the task of writing software that can run efficiently on such systems.

"Until now, the power of high-performance and parallel computing has largely been available to a limited subset of customers due to the complexity of environments and applications, as well as the challenges of parallel programming," Microsoft senior director Vince Mendillo said in a statement.

As for the new version of HPC Server, Microsoft said it offers the ability out-of-the-box to support clusters of up to 1,000 nodes as well as diskless boot and improved management and diagnostics abilities.

[ewelch]

So, I hear Vista runs faster on these machines too.

[jpckmg]

Finally. Now I can turn on automatic recalculation.

[kojacked]

LOL!

[zeroplane]

What?<br />I swear I thought this was an article from "The Onion".. Who in their right mind would use Excel on a clustered server environment as a data and analytics package? Oh that's right thoughts who drank the coolaid long ago.<br /><br />"compete against Linux in the market for compute clusters--high-performance systems built by linking together large numbers of standard servers"<br /><br />Instead of using supercomputers to calculate your excel spreadsheets how about use a free database and a simple set of free reporting classes (python, php, perl, ruby). I bet you would use less hardware, bandwidth, and electricity. Oh yeah and not have to pay the $100,000 in licensing fees..

[DevStar]

zeroplane, clearly you don't work in the finance sector. What's crazy is that there are companies that have already spent millions building out addins for Excel that their analysts can use to access their cluster backends. And I don't mean one company, but multiple companies have done similar projects. I actually worked on a project to do something simliar except we sent out the data to idle computers (remember that technology?!). <br /> <br />Excel is the workbench. People have hundreds of thousands of lines of macros written in Excel to do everyting from formatting to custom actions when hitting OLAP stores, etc... These people aren't going to change because you have this python utility that you cobbled together that doesn't work with any other asset they have. <br /> <br />And regarding power and cost. The computation is being run on the cluster. The cost of the computation on the cluster is likely not much different (and the Excel calc engine is pretty optimized, so if they're running that on the server, that likely beats the code that you're running). The cost of transferring the data back and forth is probably the smallest portion (since usually this type of computation only makes sense when the communication:computation ratio is low).

[faellen-]

I wouldn't be so shocked about Excel. Honestly, it can be a very powerful tool for computation for scientists.<br /><br />I'm not too knowledgeable about resource consumption when it comes to languages, but I was under the impression that interpreted languages like all four of the ones you listed would perform much slower than compiled languages like C. The fact they are interpreted might also deter parallelism. In Excel's defense, each entry in the spreadsheet could possibly be computed independently which happens to be something that lends easily to being computed over a distributed system.<br /><br />Bandwidth is not important if it isn't being used optimally. If you are only extracting useful information from a small portion of the bandwidth, then it is just being wasted. There is a reason that a bigger bandwidth hasn't translated to better performance in all cases.

[skratchbiker]

Hey Zeroplane, why don't you try to convince a bunch of business users to to drop their Excel spreadsheets that they've built all their business models in for python, perl, etc with an open source database? It won't happen. Excel, as much of a big pain in the rear as it is, is what business is run on. It allows those with the business knowledge to build tools that perform the complex calculations and automate their processes. They won't learn programming. Sometimes you have to take the water to the horse.

[rcrusoe]

1,000 node Windows Cluster! Just thinking about having to maintain one of those makes my head hurt. <br /><br />Convected felons would probably get to see their families more often than the sysadmins for that. :)

[timber2005]

Actually once you've got them all configured and booted, it's really effortless. There are a few interesting articles out there on it and I've worked on one, but really its the sporatic hardware issue, and more often than not its something NOT related to the cluster machines, but like maybe a router or switch not acting correctly. And if one fails, all is not lost. You've just lost 1/nth of your compuing power. And if you've lost 1/1000th of your computing power.... you've only really added 1/1000th of processing time needed. Its not all that bad :)

[fdunn3]

Very Cool Microsoft!