I’m currently working on a book about the Virginia Tech Apple Supercomputer Cluster, so when I saw a recent c|net commentary on the cluster project entitled, “A grain of salt with your Apple,” it really got my goat. Three things really bother me about poorly written commentary like this:
1. Commentary is fine, but base it on facts and be sure to get your facts straight.
2. If you don’t know what
your you’re talking about, you shouldn’t publish your opinion as if it were news.
3. If you are going to publish commentary, at least have the guts to put a name(s) to it.
The first thing in the piece to catch my eye was this statement:
“But when it comes to cluster supercomputers, an important technology expected to become the foundation of utility computing, there are hidden costs aplenty. It cost $5.2 million to buy the Virginia Tech gear, but that figure doesn’t include what the school says were “hundreds of volunteered hours of Virginia Tech faculty, staff and students to help set up the 19.25 tons of computers, routers and other equipment.”
Without actually doing any fact checking, the implication here is that if the University had strictly paid for all of the labor incurred when building this project, the costs would have significantly increased. Therefore there are enormous amounts of hidden labor costs that affect the bottom line.
First of all, the $5.2 million also included
construction costs (actually construction costs for power and cooling upgrades, were an additional $2 million - however, those are not platform specific, so whatever vendor they selected, those costs would be the same), not just the “gear.” And while there was a large body of volunteers working on the project, their role consisted primarily in removing 1,100 G5 computers from the box, seating a special network card, powering the machine to test for DOA, and helping to place the machines on the specially constructed racks. The cabling and setup of routers and other equipment was primarily done by University Staff and representatives from hardware manufacturers. In fact, while I was there documenting the project, I saw the University staff (who are on salary) on their hands and knees running miles of cable. I’d also like to point out that the reason such a large body of volunteers was required was not because they wanted to exploit a free workforce, but because the University was on a very tight deadline in order to make the Supercomputer list.
When you work out the details (and you’re very generous), you might add an additional $400,000 in hidden labor costs, but this doesn’t really diminish the accomplishment. Even at $5.6 million it would still have been the best price performance leader on the TOP500 List in history!
The next thing to catch my eye:
“In an academic environment, there are plenty of graduate students on hand to figure out the best arrangement of processors, memory and network gear for a given task. Students also can translate software written for other computers to Apple’s systems, which with a single machine now on the top 500 list are far from prevalent in the supercomputing arena.”
Sorry, but this information is not just wrong, but completely disrespectful to the highly skilled project leaders like
Dr. Srinidhi Varadarajan, an assistant professor of computer science at Virginia Tech, and Jason Lockhart, director of the College of Engineering’s High Performance Computing and Technology Innovation, who initiated the project. Come on folks, you don’t let grad students design critical systems. Dr. Varadarajan did most of the coding, along with the Mellanox team that wrote the InfiniBand driver. And while the Virginia Tech cluster is the first Mac on the list, it is in good company as one of the many Unix systems on the same list. And I’d also like to point out that all of the software used on the project was open source…yes all.
“So the Apple project at Virginia Tech may be a wonderful educational project, but commercial customers who have less interest in experimentation are more likely to pay specialists at Linux Networx, RLX Technologies, IBM, Sun Microsystems, Dell or Hewlett-Packard to plan the plumbing, package the software and plug in the cables. And those companies aren’t going to rely on Macs.”
Experimentation? With a quickly approaching deadline, and millions of University dollars on the line, there was no room for experimentation. This had to work and had to work the first time, mistakes or miscalculations would be devastating. The folks at Virginia Tech are specialists. The Mac cluster isn’t their only cluster and Dr. Varadarajan is a respected expert in his field. Why this assumption that the experts from Apple, Mellanox, Virginia Tech and others didn’t know what they were doing and only folks from Dell (of all people) are the only experts who can build these solutions?
The c|net piece was nothing but spurious speculation and Mac bashing from a publisher who has a history of negative press about the Apple platform. Shame on anyone who assumes that this solution cannot be duplicated again with less money than competitive solutions, and shame on c|net for implying that this was all a worthless exercise and corporate clients would be smart to seek their solutions from a more reliable hardware company.
If corporate clients were smart they’d pick up the phone and call Virginia Tech and Apple computer who have the third fastest supercomputer on the Top 500 list, and ignore c|net, which to my knowledge, didn’t make the list again this year.