Life science organizations looking for high-performance computing (HPC) equipment to meet their computational needs had many varied options in 2004.
Last year, the major trends in HPC equipment included an emphasis by systems vendors on equipment integration and, naturally, the availability of increasingly more processing power in more densely packed form factors.
On the server side of things, Microways Navion-8, an eight-way SMP system that can support up to eight AMD Opteron 800 Series processors and up to 128 GB of memory, is typical of the incredible processing power that can be packed into a rack-mounted single server.
When it comes to desktop systems, dual-processor units from virtually all the systems vendors deliver processing power that, a few years ago, would classify these systems as workstations.
On the application support side of things, new systems designed specifically to deal with the computing requirements for visualization started to emerge. These systems took advantage of the increasingly more powerful graphics processing units (GPUs) that became available in 2004. Many of the new systems aimed at visualization applications also leveraged the 64-bit processors' ability to support very large data sets. Most of the products introduced in this arena were desktop systems. However, some systems, such as SGI's new Prism line, offered a modular architecture that could scale to support thousands of processors and very large amounts of memory.
In a different type of application, Orion Multisystems introduced 12- and 96-node workstations that essentially are desktop and desk-side clusters. The systems use tightly integrated, low-power-consumption processors to give researchers the computational power of a cluster that is easy to operate and manage. These systems can be used as a normal cluster, but one use envisioned for these products is as an application development tool for researchers that would let them test a clustered algorithm before running it on an operational or production cluster.
These products were just the tip of the iceberg. What follows is a compilation of the major life science HPC trends for 2004.
In a continuing trend from 2003, 2004 saw most systems vendors offering pre-configured systems of all sizes that tried to make it easier to use an HPC cluster. Vendors including HP, IBM, Sun Microsystems, SGI, Dell, Linux Networx, Microway, and Appro all offered systems that included the server hardware, pre-installed operation system software, and in many cases a slew of open source informatics algorithms.
The offerings ranged from 1000-node systems for large research organizations to smaller systems for small labs. One interesting 2004 trend was that the systems vendors put a lot of effort into making the lower end of their product lines easier to install, administer, and manage.
For instance, in March, Sun introduced the Sun Fire Starter Cluster for Bioinformatics, a 19-inch rack on rollers. The system, a smaller version of an existing, Sun pre-bundled informatics system, was designed for small labs and departments within larger companies. Essentially, the Starter Cluster was aimed at scientists who need a certain level of computational power, but who did not necessarily have an IT staff around to maintain an HPC system.
In the second half of 2004, a new integration trend started to emerge. Some systems vendors, such as Appro, HP, RLX Technologies, and others, started to integrate high-performance internetworking switches within their racks and blade systems.
Typically, switch integration is done either by a third-party systems integrator, a value-added reseller, or by a life science companys IT department. One benefit when a systems vendor takes over this job and integrates the switch is that a single management system supports the server compute nodes and the switch. As a result, the switch is often easier to manage.
When it came to the high end of the HPC market, IBM's Blue Gene program, which has protein folding as its showcase application, reached some notable milestones.
In one year's time (from November 2003 to November 2004), IBM was able to scale Blue Gene''s processing power significantly. In fact, in November 2004 two Blue Gene working prototype systems placed first and eighth on the list of the world's Top500 supercomputers. In contrast, a single Blue Gene working prototype placed (a still-respectable) 73rd one year earlier.
The number one system, called the Blue Gene/L DD2 beta-System, achieved a sustained Linpack performance benchmark of 70.72 TeraFLOPS (70.72 trillion operations per second) and a peak performance of 91.75 TeraFLOPS. In contrast, the system that placed 73rd in November 2003 had a sustained performance of 1.4 TeraFLOPS and a peak performance of 2 TeraFLOPS.
The first commercial Blue Gene system, dubbed Blue Gene/L, will become available this year. And IBM already has a handful of paying customers lined up, including the Computational Biology Research Center (CBRC) of Japans National Institute of Advanced Industrial Science and Technology (AIST). The CDRC will use a Blue Gene/L system to conduct 3D protein folding simulations aimed at accelerating research into new drugs. (Other non-life science customers to date include Argonne National Laboratory, the Dutch astronomical organization ASTRON, and the Lawrence Livermore National Laboratory.)
Several other life science systems vendors also placed systems in the top 20 on the Top500 list of the world's most powerful supercomputers. For example, an SGI Altix system placed second, and HP, Dell, Linux Networx, and Fujitsu all have systems in the top 20. (Most of these systems are not used for life science research, but systems with the same architecture from these vendors are widely used within life science organizations today.)
Desktop and Workstation News
Systems vendors also delivered more processing power to the individual user in 2004.
Apple, Dell, HP, IBM, SGI, and Sun Microsystems introduced single- and dual-processor systems based on 64-bit Itanium, Opteron, or PowerPC G5 chips and running Linux, Windows, or Mac OS X operating systems.
Additionally, there new systems were rolled out based on Intels Extended Memory 64 Technology (EM64T), which allows a 32-bit processor to run 32- and 64-bit applications. Systems that use EM64T technology can support more memory (and therefore, larger data sets) than traditional 32-bit systems.
With both the 64-bit and EM64T-based systems, the combination of greater processing power and support for more memory makes such systems suitable for some applications that in the past were only able to run on dedicated RISC-based systems.
The main application areas for these new desktop systems are computational biology of large databases, and visualization.
Business-wise, two stories stood out in 2004: Paracel's demise and RLX Technologies decision to focus on system and server management software.
In late December, blade computing pioneer RLX Technologies announced it would exit the hardware business to focus its resources on server and cluster management. The company plans to offer its Control Tower management software for the management of the major HPC systems on the market.
In September, life science HPC vendor Paracel ceased operations. The company sold systems that were designed to speed informatics and drug discovery research. Notable among its products was the GeneMatcher2, a massively parallel supercomputer based on application-specific integrated circuits (ASICs).
Despite Paracel's demise, systems vendors did not abandon the market for tightly integrated systems that include dedicated hardware to accelerate the execution of informatics applications.
For instance, early in 2004, Cray acquired systems vendor startup OctigaBay, who was developing an Opteron-based system that would run Linux, use an architecture that reduced PCI and memory contention bottlenecks, and that could use Field Programmable Gate Array (FPGA) technology to accelerate the execution of many common informatics applications. Cray launched the system as the XD1 supercomputer.
Other companies that offer hardware acceleration technology in 2004 included TimeLogic, Star Bridge Systems, and Fujitsu. (For more details about trends in this segment of the HPC market, see Accelerator Market: Not Dead Yet.