64 Bit Blog

Bobby Fischer 1943-2008

When every blog says, "Here's my two bits,"
It seems a stretch to speak of 64.

]]> Here's my two bits, then, in 64 words.
Once the Tiger Woods of the 64-square madhouse,

Anti-semitic Jew, anti-American American hero, grandmaster Bobby Fisher
Has died at the fitting age of 64.

While 64-bit computers may now challenge chess grandmasters,
None can fathom the mind of a man.

Why I love writing online: Reason 99

Of the many things I love about electronic publishing is that I will never be compelled to atone for a bad prediction by running my words through a blender and drinking the resulting text-smoothie in front of hundreds of people.

]]> If you believe everything you read in the Wall Street Journal, you should be worried about the coming "exaflood" that will soon drown the last mile of internet connectivity and for which the only solution is to give up on this idea of net neutrality.

As Nate Anderson explains in one of those long essays at Ars Technica, the exaflood threat may just a wee bit overblown. Not too surprising, considering where the author of that WSJ scare story works. And a scary story it is: he's gone and got the Hollywood Reporter all stirred up.

Thanks to Nate for reminding us that Bob Metcalfe, who knows a lot more about network capacity than Bret Swanson, had to eat his column on stage a decade ago for predicting internet "gigalapses" in 1996. "I was wrong," Metcalfe wrote. "I ate the column. I am sorry. I am not worthy."

Perhaps we can save Swanson the trouble (and indigestion) and say the same for him.

The Apple of their eye

In a long series on Apple's encroachments into what we are now supposed to call 'the enterprise,' Andrew Burger at MacWorldNews takes time out to talk about how and why Apple is so popular in scientific computing.

]]> The example he cites is Harvard's Initiative in Innovative Computing (IIC), which was set up in 2006 and attempts to foster the use of innovative computing tools in scientific research. "The IIC enables science that would not be possible today without the innovative use of computational technology," IIC asserts.

What Burger asserts is that nearly every researcher in the interdisciplinary IIC program chose a Mac as their desktop computer. One reason, Berger figures, is the strength of Apple's 3-D imaging and visualization tools. But the Windows compatibility, Unix functionality, and the Mac GUI are probably part of the reason for Mac's popularity among scientists.

Or maybe they just like the commercials. Ho ho ho.

Google trawls for knols

Udi Manber has quite a sense of humor.

Google's VP of Engineering wonders on his personal website, "Why is there no synonym for 'thesaurus'?" I wonder why someone so interested in online content hasn't substantially updated his personal website in ten years.

]]> Manber's interest in web-based content is what this posting is actually about.

Google generally doesn't do dumb things, but this project to develop 'knols,' which are basically encyclopedia entries, has me scratching my head. And not just because 'knol' is such a butt-ugly word. After all, 'blog' is pretty grotesque, and there are a grotesquely huge number of those things.

But what's this business about people who post to the Web not getting enough credit? Certainly not true of the blog-sotted web I frequent, although I suspect that this is just a dig at Wikipedia. As is the idea that static knols will compete for attention, as opposed to the Wikipedia model of refining one entry to perfection. Snark tag on that last word.

Out on that bloggy web where people are sharing their Google-unauthorized knols about this Google project, it's being described as a terrible idea and a step too far. The phrases 'grassy knol' and 'death knol' recur. Andrew Orlowski thinks that for Google to attack Wikipedia is ingratitude. Rick Munarriz gets it that Wikipedia works because it is understood to be inaccurate by nature. If Googlepedia promotes the reliability of its experts but is no more reliable....

Well, whatever the Google strategic thinking is, it's apparently keeping someone awake nights, judging from the sample knol Google provides. (It's written by Manber's wife: make of that what you will.)

Doubtless the whole thing will be made clear soon. I'll keep checking Manber's Wikipedia page for the real skinny.

Tapping into Unstructured Data

Bill Inmon coined the term 'data warehousing,' wrote the first book on the subject, and held the first conference on data warehousing. Lately he's turned his attention to the broader challenge of managing unstructured textual data.

]]> His new book Tapping into Unstructured Data [with Anthony Nesavich, Prentice Hall, 2008] tackles the challenge of integrating such messy data into business intelligence.

If you think like a database, and when you're compiling and processing and using corporate business intelligence, you pretty much have to think like a database, unstructured data doesn't even exist. There's information there, of course, but it is not expressed in a form that can make any meaningful connection with databases. It is as though the databases are blind to it.

How you cure that blindness is the subject of the book, and for me the most enlightening part of the book was the case studies. Here the authors paint a clear picture of the iterative process of extracting meaning from unstructured data, using the meaning extracted at each step to inform the next step.

Vast amounts of information are tied up in unstructured documents. This blog is adding to the pile. It is really encouraging that efforts like this book are underway to extract from it the kind of meaning that is needed for business, medical, and other decisions.

Apple's ZFS

It's Sun's ZFS, of course, but I've been tracking news of Apple's implementation of Sun's Zettabyte File System, and I'm not the only one.

]]> French blogger Blaese thinks the main limitation now is the UI: that you have to access it as root from the command line. He also thinks that there's some confusion in Appleland over pools and volumes and disks.

I wonder if Sun didn't sow seeds of confusion by naming it the Zettabyte File System when it's not a file system. That's what Tom Yaeger said when he called it "nearly perfect" and said it was "limitlessly flexible" and "easy enough for a child to manage."

Even before Leopard was released, Gregg Keizer was speculating that ZFS could be Apple's secret weapon. Of course that was shortly after Jonathan Schwartz misled everyone about what Apple was going to do with ZFS. Gregg pointed out back then that Sun was already apparently regretting the name and, IBM-like, asking people to treat ZFS as a name and not an acronym.

Schwartz thinks ZFS is a big deal. One company that agrees, but considers it a big legal deal, is NetApp, which has sued Sun, claiming that ZFS rips off NetApp's patented technology. Sigh.

ZFS is indeed a big deal. If you don't know all about it, check out Sun's introduction. And for Apple's implementation, you might start with the man page.

Coming to Reno

Here are a few more of the folks exhibiting or presenting at SC07, the ACM/IEEE-sponsored International Conference for High Performance Computing, Networking, Storage and Analysis, running from November 10th through the 16th at the Reno-Sparks Convention Center.

]]> RapidMind will be showing off the latest version of its software designed to let you actually benefit from the added processing power of multi-core chips.

Advanced Cluster Systems will be making an announcement on Tuesday that I think will be interesting. The provocative phrase 'move beyond ridiculously parallel solutions' will appear in it.

NEC just announced what they claim is 'the world's fastest vector computer,' and some of the key people involved with that will be at SC07. And DataDirect is debuting what they call 'the fastest storage system on the planet.'

Not for nothing do they call it SUPERcomputing 07.

SC07 Update

SC07, the ACM/IEE-sponsored International Conference for High Performance Computing, Networking, Storage and Analysis, runs from November 10th, that's tomorrow, through the 16th at the Reno-Sparks Convention Center. Here are some of the companies and products I know of that will be appearing at SC07.

]]> Calgary-based Acceleware will be showing off their Accelerator Board and ClusterInABox Workstation solutions at the NVIDIA booth; tools they claim put supercomputing capabilities on the desktop.

IBM will host an invitation-only event to meet and chat with HPC execs from the company, like Michael Henesey, the Vice President of IBM's Deep Computing. Tony Belfi will speak on 'Innovation Beyond Imagination: The Road to Petaflops Computing,' and IBM will also do demos related to Blue Gene/P.

Cray will be there with news of launch of the Cray XT5 family of supercomputers, including the industry�s first integrated hybrid supercomputer.

Infinera (INFN), one of the companies behind Internet2's high-speed network for the
scientific community, will be making three announcements at the show.

NASA will be there, talking about their climate work in addition to the obvious space stuff. The DOE will be there.

Interactive Supercomputing will announce a new version of their parallel computing platform, Star-P, around the show.

Dell will be talking up their cluster and cloud computing initatives.

A new book that sounds pretty interesting, Petascale Computing: Algorithms and Applications, edited by David A. Bader, will be released at the show.

Sun gets smart about music recommendation

I grew up on "American Bandstand," and I used to cringe when it came to the Rate a Record segment, because I knew that the kids were going to give the record a totally meaningless numerical rating and when Dick Clark asked why they gave it an 87 or whatever it was, they would invariably say, "It has a good beat and I can dance to it."

]]> Online music recommendation systems built into iTunes and Amazon are more sophisticated than this, giving you the benefit of the musical judgement of others who are in some sense like you, but ultimately they amount to the same thing: the essentially unexplained preferences of strangers.

Sun Microsystems has a different idea with its project Seach Inside the Music. It uses machine learning to actually analyze aspects of the music you like. Sun plans to release the technology as open source, maybe within the next six months.

HPC in Reno

Next month, supercomputing folks from all over the world will be converging on northwest Nevada for a major high performance computing conference. I'll be there. Will you?

]]> SC07 is the ACM/IEE-sponsored International Conference for High Performance Computing, Networking, Storage and Analysis. It runs from November 10th through the 16th at the Reno-Sparks Convention Center. Keynote speaker: Deputy Raineesha Williams of the Reno Sheriff's office.

Just kidding. Although all the officers of Reno 911 qualify as high performance. SC07 includes a Disruptive Technologies track. Raineesha's on it, you can bet.

The keynote will be delivered by Neil Gershenfeld, Director of the Center for Bits and Atoms, MIT. I want to hear him tell me how I'll soon have an inexpensive printer on my desk that will print functioning 3D artifacts.

The Simplest Turing Machine Has Been Proven Universal

This spring, Stephen Wolfram offered $25,000 for the first person to prove that a very simple Turing machine has the property of being universal. A 20 year old has won the prize.

]]> Wolfram is the Macarthur Fellow genius who wrote Mathematica and founded Wolfram Research, and who, in his mammoth book A New Kind of Science (NKS) introduced what indeed looks like an altogether new kind of science.

This latter work arose from the discovery that you can get what seems to be unbounded complexity out of systems that seem far too simple, such as a few bits of data and a program that can be described in a few words.

In NKS, Wolfram shows that many different kinds of simple systems are essentially equivalent and states his Principle of Computational Equivalence: that almost all processes that are not obviously simple can be viewed as computations of equivalent sophistication.

Which is important because... but what am I doing? I'm not going to try to explain Wolfram's theories. I'm just here to report that on October 24, the winner of this contest will be announced, and we will all see that a Turing machine with just 2 states and 3 colors is a universal computer.

Tweaking C for mesh-based physics code

Jeff Keasler dropped me a line to let me know about a C language extension he's written for mesh-based physics coding. Stuff like finite differences.

]]> He wrote this HPC extension in under 500 lines of ROSE code. The basic idea, he explained, is to "create a single schema file for your program that describes the topological realtionship among fields being used. The schema file is then used to help compile your C code written in 'vector' notation. The compiler takes the 'vector C' code and applies the schema to create either structs or arrays in your compiled code."

It's a C extension, so it'll work alongside your existing C code, and can generate up to a 2X performance improvement, Jeff says. At his website he provides some working examples to show how you use it.

That's a nice thing about being an editor: not all the unsolicited email is spam. Some of it is worth sharing.

Be a super(computing) fellow

Budget surplus: there's a phrase you don't hear much these days. Even rarer: they want to know whether they should give some of it to you.

]]> The Steering Committee of the ACM/IEEE SC (supercomputing, that is) conference have decided to steer the surplus funds from their '05 conference to PhD fellowships in high-performance computing.

Pointing out that the National Research Council study on Supercomputing recommends "[f]unding models that encourage and support the education of the next generation, as well as those that provide the supercomputing infrastructure needed for their education," the ACM and IEEE Computer Society has set up a fellowship program, and are looking for candidates right now.

The deadline for applying for the 2007 award is the end of this month (the 30th, actually), so if you're in a PhD program and interested in HPC, you might want to jump on this.

Fortran, Numerics, and MacOS: Strange Bedfellows?

The answer to that question is: Not really. You might not think of Fortran when you think of modern programming, and modern operating systems. But despite it not being the flavor of the moment, there are good reasons not to count Fortran out. In addition to there being huge codebases already in existence that no one feels any need to rewrite, Fortran is still quite a strong choice for high performance computing and numerics. Jon Erickson talks to Wood Lotz, president of Absoft, about the IMSL Fortran Numerical Library, recently ported to MacOS.

RapidMind In Depth

I know we've talked about RapidMind before in this space. (Jon Erickson recently interviewed RapidMind's chief scientist, Michael McCool.) But there's a lot more to say. McCool and RapidMind founder Stefanus Du Toit take us on a technical tour through coding an app for multicore processors using RapidMind. The basic premise? Instead of spending your time optimizing your algorithms for multicore, or learning custom languages or environments, let your tools worry about the multithreading.