As the deadline for final orders of HP’s AlphaServer systems approaches, a number of the company’s life science customers view the phase-out of the platform as an opportunity to remake their IT infrastructures with lower-cost commodity architectures.
Last month, Sun announced that the Institute for Genomic Research — a longtime Alpha customer — had replaced its cluster of 15 HP Alpha servers with three Opteron-based Sun Fire x64 servers running Linux.
Vadim Sapiro, director of IT at TIGR, said in a statement that the six-year old Alpha servers were growing “sluggish and unreliable” and required “massive power and cooling” systems that were extremely costly to operate.
The three Sun Fire servers provide the same level of performance as the 15 Alphas, Sapiro said, while reducing TIGR’s power and cooling costs by more than 70 percent. The initial purchase price was also considerably lower, at $30,000 per Sun server compared to the $100,000 TIGR paid for each of its Alphas in 1999.
The Wellcome Trust Sanger Institute, another longstanding Alpha customer, is “pretty much in the same position” as TIGR, Phil Butcher, head of IT at the Sanger Institute, told BioInform this week.
Butcher noted that the Alphas — originally sold by Digital Equipment and later by Compaq — “were the cornerstone of our IT provision” at the height of the Human Genome Project. At the time, the Alpha was the undisputed platform of choice among high-powered bioinformatics users.
In addition to TIGR and Sanger, Compaq claimed Celera Genomics, the Whitehead Institute (now the Broad), Genentech, Entelos, Genoscope, and a number of other high-profile life science organizations among its customer base.
But in 2001, Compaq decided to retire the Alpha platform and standardize on Intel’s Itanium chip [BioInform 07-09-01
]. The phase-out officially ends on Oct. 27, which is the last day customers can place orders for AlphaServers.
HP, which acquired Compaq in 2002, has pledged to support the Alpha systems through 2011 and recommends its Itanium-based Integrity servers for those customers looking to retire older Alpha systems, but it appears that some customers are seeking other options.
“We’re now in [the] process ourselves, in fact, of going through the same process as TIGR — of moving onto large-scale enterprise Linux clusters to replace our Alpha Tru64 clusters,” Butcher said. “HP has decided not to develop the Alpha chip any further, and of course now with the advent of fast commodity chips, dual-core processors, and so on, there’s a much more cost-effective price/performance point [available] for us.”
Sanger currently runs a mix of systems from HP, IBM, SGI, and other vendors. Butcher said that he’s been particularly pleased with AMD’s Opteron chip because it works well with both 32-bit and 64-bit code.
“We see our future with commodity computing,” he said.
HP officials were unable to comment before press time.
The Sun Also Rises
For hardware vendors like Sun, the Alpha phase-out represents an opportunity to gain market share in the life science high-performance computing sector.
Joerg Schwarz, director of health and life sciences at Sun, noted that HP inherited a large installed base of life science customers when it acquired Compaq, and with the “life cycle of these servers coming to an end, there’s really no future for Alphas, so a lot of life science institutions are facing the same migration challenges right now.”
Nevertheless, Schwarz said, HP is still a “viable competitor in this space,” largely due to its installed customer base. Brand loyalty can pose a challenge in migrating customers to a new computing platform, he noted. “If you’ve had good relationships with your vendor over several years, there’s usually no need to change that relationship unless there are some advantages or other points that convince you.”
“HP has decided not to develop the Alpha chip any further, and of course now with the advent of fast commodity chips, dual-core processors, and so on, there’s a much more cost-effective price/performance point [available] for us.”
Schwarz said that Sun’s primary advantages over competitors like HP, IBM, and Dell include the “most advanced Opteron server line in the industry,” as well as a willingness to buck one of the most common traditions in high-performance computing — vendor lock-in.
“The tradition in HPC is that you always try to have some proprietary part that creates stickiness between the customer and the vendor,” Schwarz said. “For example, if you have a Cray vector supercomputer, it’s very, very hard for you to switch vendors because you have to rewrite your whole application because nobody else makes a system like that.”
Sun, however, offers “zero cost of exit,” Schwarz said. “If you go with Sun technology, you get a lot of benefits, but if at one point in time you decide to buy a compute platform from another vendor, you don’t have to change your entire infrastructure.” Unlike the AIX and Tru65 operating systems, which only run on systems from IBM and HP, respectively, Sun’s Solaris can run on any Opteron system, Schwarz said.
So how does the company plan to compete against other companies selling commodity Opteron boxes? “We put a lot of engineering time and effort into designing these systems for use in this kind of environment,” Schwarz said. “It’s not just a cheap box with a cheap motherboard where a lot of the design and engineering elements come from the desktop or the gaming market. It really has been designed for deployment in a cluster environment.”
The TIGR agreement should give Sun the opportunity to prove that this is indeed the case, and that the company is a contender in the life science HPC market.
“We think if we work with TIGR and show that TIGR can improve their efficiency, their throughput, and other factors because they work closely with us, then other institutions basically can use that as a template and utilize these benefits in their own organization,” Schwarz said.