Washington–Supercomputers keep breaking records for processing speed, but software to operate them has not kept up with that increasingly zippy hardware. The often-rickety supercomputing computer code is becoming an obstacle to making better weather models, medical simulations, and other applications of high-performance computers, said experts at a conference here Wednesday on the future of academic supercomputing.
“Codes are still being used from the 1960s,” said Ed Seidel, director of the National Science Foundation’s office of cyberinfrastructure, in an interview at the meeting. “Those have to be retooled or rethought” to take full advantage of the latest supercomputers, he said.
Attendees at the meeting said one of the most popular computer languages used to create programs for supercomputers is Fortran, which went out of style among conventional programmers decades ago and is rarely even taught in college computer-science departments today. It’s as if your new laptop still ran MS-DOS, the operating system that predated Windows on personal computers.
In other cases, college researchers run their supercomputing projects using homemade software which was written decades ago, but which has been patched to run on the latest machines, said Phil Smith, senior director of the Texas Tech High Performance Computing Center at Texas Tech University. “The people who wrote those codes have gone away — they’ve retired,” he said.
The issue was one of the several raised at the two-day meeting, which celebrated the 20th anniversary of the Coalition for Academic Scientific Computation.
The problem is that it is hard to find money to support the development of a new breed of supercomputing software, several attendees said. The National Science Foundation has long helped colleges buy supercomputers, but gives few grants to build software for them.
Representatives from supercomputing manufacturers who attended the meeting said there is not a large-enough market for them to focus on software. Peter Ungaro, president and chief executive of Cray Inc, one of the largest supercomputer manufacturers, said he hopes that university researchers can jointly develop a new generation of software that would run on any company’s hardware.
Mr. Seidel said the National Science Foundation recently set up a committee to look into the supercomputing-software issue and make recommendations for how to fix it. Those recommendations are expected within the next year, he said.
But some attendees said they have little hope that the committee will solve the problem. “I wouldn’t even put ‘hope’ in the same sentence as the committee,” said Jess Cannata, head of high-performance computing at Georgetown University. “It’s so hard of a problem that no one wants to touch it.”
11 Responses to Supercomputers Often Run Outdated Software
benjamin_geer - September 23, 2009 at 7:46 pm
This is a baffling article. Most of the world’s fastest supercomputers (including the ones made by Cray) run some version of the Linux operating system, which is anything but outdated:http://blogs.computerworld.com/the_worlds_fastest_computers_are_linux_computersTo be specific, 77% of the fastest 500 computers in the world run Linux:http://www.top500.org/stats/list/33/osThat's hard to square with the claims made in this article.
wbgleason - September 23, 2009 at 10:05 pm
Not sure what to make of this either…Rickety supercomputer code? Care to be a bit more specific? Most of the folks at our place – the Minnesota Supercomputer Institute – have pretty good custom codes for their stuff where appropriate.For anyone who cares there are compilers/analysis tools that allow one to squeeze a lot out of code. These can be used to isolate the 10% of the code that needs serious optimization. Most folks who use a lot of supercomputer time know about this. I’d think for this to be a legitimate article that parallelism would at least be mentioned. And even this problem seems to be getting attacked pretty well. The NAMD code developed at Illinois scales pretty well. Although I wouldn’t start a new student out with Fortran anymore, Fortran has its place. The codes of George Sheldrick – originally written in the sixties but under constant developement – were and are written in Fortran and they are pretty damned efficient. To conclude: this article is a little disappointing.
11246158 - September 24, 2009 at 4:54 pm
Benjamin, you are correct that the OS is not the issue — it’s about the modeling and simulation codes used by researchers, many of which are community developed codes (not commodity codes) and which have evolved over many years. As processors becoming increasingly parallel (multicore and manycore), many codes will need to be tweaked or outright overhauled in order to take full advantage of the more powerful computing hardware that is available. This is “unsexy” work and there if definitely a shortage of grant funding to support it, so I am encouraged that Ed Seidel and other’s at NSF are recognizing and addressing the issue.
jdxxxe - September 24, 2009 at 9:17 pm
Hey, FORTRAN programming was FUN! It was relatively easy to learn enough to put togerher programs that actually did interesting things. And it was generally relatively easy to follow what a program was doing, due to its structure. Now here in my old age perhaps I can hire myself out as a FORTRAN expert — having managed to successfully bypass about four or five generations of subsequent programming languages — I’m uncontaminated by any of that nasty object-oriented stuff. The income from FORTRAN consulting, combined with my Social Security, the tatters of my TIAA-CREF funds, and the steady income from my tin cup of pencils ought to be enough to keep me in bourbon and dogfood through my Golden Years…
scottmidkiff - September 25, 2009 at 9:45 am
The problem, as has been recognized in some of the comments, is with the application domain code and with domain-focused middleware and libraries. There are some domains where robust funding has allowed the community to keep up with the rapid changes in underlying architecture. However, many domains have not been able to keep up and end up using old codes on new hardware, or simply can’t leverage the new hardware. A concern is that new computing facilities are serving a smaller and smaller set of science and engineering domains. Ed Seidel is focused on the right things.
11224067 - September 25, 2009 at 11:41 am
This article is very strange. Anyone who thinks that FORTRAN is outdated obviously does not know about the Absoft Corporation in Troy Michigan and the website is http://www.absoft.com. The comment would be better suited to COBOL, however, I don’t see that software running on a supercomputer.
saltydog - September 25, 2009 at 3:29 pm
And the person who mentioned COBOL above obviously does not know about MicroFocus, nor that many, many large federal agencies still run large IBM mainframe shops with zillions of lines of COBOL, and are also still doing new development using COBOL. Please don’t misunderstand … COBOL _is_ outdated, and those agencies are pretty much the lands that time forgot. But it’s not outdated because of who does/doesn’t sell it or use it. I can probably still buy vaccuum tubes somewhere, but that doesn’t make them modern. Same with FORTRAN (and COBOL). Besides, which language to use is a religious argument nobody can win. Use what works to solve your problem!
michaelleerilee - September 25, 2009 at 4:09 pm
@wbgleason “have pretty good custom codes for their stuff”This is part of the problem. There is not a very strong division of labor in scientific computing, though it has gotten better over the years as the market has broadened. Doing cutting edge science on national resource class machines entails custom fitting low serial number or prototype software on low serial number or prototype machines with low serial number or prototype tools. Periods of architectural stability allow the advancement of tools and techniques (productivity) until the next break in capability. In a world where customization is king and the technological rug gets pulled out from under you programming to the lowest common denominator becomes the norm.The NSF has to figure out what its goals are and where it wants to spend its money. Funding to scientists will go to generate science which will push computing technique “as needed” in an “ad hoc” way. Funding the computer scientist will tend to push computing technique but the benefits of this may only diffuse out through the scientific supercomputing community as the general level of computing technology advances. Maybe there are specific projects that may be of more immediate impact, but I’m not hopeful. There is also a semantic mismatch or impedance gap between the two communities.
rchrd - September 28, 2009 at 12:52 pm
Statements about the demise of Fortran are greatly exaggerated. Fortran is still used by a majority of high performance technical computing applications and that is because it is a mature language with mature compilers that know how to generate highly optimized and parallelized code.Fortran (not ‘FORTRAN’) is the most successful programming language and in many ways the most appropriate language for easily coding numerical algorithms. The latest version of the Fortran specification describes a language far advanced from the original FORTRAN from 1966. Fortran lives! Long live Fortran.
laoshi - September 30, 2009 at 11:59 am
“It’s as if your new laptop still ran MS-DOS, the operating system that predated Windows on personal computers.”Actually, Microsoft’s disk operating system still works under the Windows shell. And FORTRAN is a programming language, not an operating system.It’s as if Jeff Young doesn’t know anything about the computer technology he opines about.
shayden69 - September 30, 2009 at 11:37 pm
Correct me if I’m wrong, but as most supercomputers are primarily number crunchers (do ANY (super)computing “benchmarks” include processing Word/Powerpoint documents?), I don’t see why Fortran (BTW, my 3rd language learnt, after assembly and Basic) should bear any fault here. Just “parallelize” Linux, optimize your hardware architecture to run math intensive applications ON Linux, and what better world to run FORmula TRANslations (and maybe even object orient it, so EVERYBODY can have fun).Regarding the statements “Representatives from supercomputing manufacturers who attended the meeting said there is not a large-enough market for them to focus on software”, WOW, what a cop out. It seems that if they had better software, they could sell even MORE hardware, to solve even BIGGER problems.And “Peter Ungaro, president and chief executive of Cray Inc, one of the largest supercomputer manufacturers, said he hopes that university researchers can jointly develop a new generation of software that would run on any company’s hardware.”. We’ve got that now, it’s called the “Java Virtual Machine”, which is just one more layer of overhead your machine’s got to deal with. That’s like saying “every computer manufacturer’s got to give us hardware that runs the same thing the next one’s got.Any as far as the MS-DOS, I thought later versions of Windows ran native, and they only kept DOS around for fossils who can’t work a mouse.