New York researchers opt for Big Blue blades
The University at Buffalo turns to IBM for a 266-blade server system to investigate basic disease mechanisms--the beginning of what it plans will be a system about 50 times larger.
- Shankland covered the tech industry for more than 25 years and was a science writer for five years before that. He has deep expertise in microprocessors, digital photography, computer hardware and software, internet standards, web technology, and more.
The IBM system will have the processing speed of about 1.2 trillion calculations per second, said Jeffrey Skolnick, director of the New York state university's Center of Excellence in Bioinformatics, who declined to say how much the system cost.
"My goal is a system with 25,000 processors in the next two to three years," Skolnick said in an interview Thursday. "This is the next step of what will be a much larger cluster."
IBM's BladeCenter chassis can accommodate as many as 14 dual-processor blades. The university is using HS20 blades, each with dual 2.8GHz Xeon processors and 1GB of memory and version 2.1 of Red Hat Enterprise Linux. It also uses an IBM FastT700 storage system with 5 terabytes of capacity, IBM said.
Blade servers are catching on but still can't match the input-output capabilities of standalone servers. IBM's blades communicate by plugging into a "backplane" that can communicate only at 1-gigabit-per-second Ethernet speeds; standalone servers with their own communications gear can take advantage of higher-speed technologies such as InfiniBand.
But Ethernet is fine for many tasks, including the University at Buffalo's, which Skolnick said don't require extensive communication between the different computing nodes.
Protein boost
The university is studying the structure of proteins, which are made of long chains of building blocks called amino acids. The amino acid sequence is stored in DNA; when proteins are constructed one amino acid at a time, the molecule folds up into a complex three-dimensional shape whose features determine the protein's functions.
Computer simulation of protein folding takes extensive horsepower. IBM's Blue Gene supercomputer project, for example, has as its goal some of the most complicated protein folding exercises.
Skolnick said the initial IBM system is able to model proteins with about 200 to 300 amino acids. With current horsepower and modeling techniques, a computer model correctly models a protein about two-thirds of the time, Skolnick said.
The university is also collaborating with IBM life science researchers for the project, using IBM pattern-recognition algorithms to see if different proteins with matching features have matching functions as well, Skolnick said.
The IBM system augments a Dell machine with 2,000 dual-processor servers, said Skolnick, who wasn't happy with his Dell relationship.
"I'm willing to accept that this is a very large installation and things go wrong," Skolnick said, but "they were not responsive in a timely manner to fix them." In addition, Dell "didn't seem especially interested in expanding the cluster" and didn't deliver a number of blade servers originally due in November 2002, he said.
Dell spokeswoman Wendy Giever said Skolnick was one of three Dell supercomputer customers at the university. The others--one of them No. 38 on the Top 500 list of the world's fastest supercomputers--have had only two server failures in two years, Giever said.
"It's frustrating that Jeff is unhappy, and we wish he wasn't. We work hard to give the best customer experience," Giever said.
The SCO Group, a former Linux company that bought some Unix assets and now argues that Linux violates its Unix intellectual property, has sent threatening letters to many Linux users. SCO targeted supercomputer users, among others.
Skolnick said his university was among those who received a letter. The researchers referred the letter to the university's legal counsel, he said but added: "It didn't scare me."