By Salvatore Salamone
June 12, 2002 | Most distributed computing systems use either clusters of back-end servers or desktop computers, but usually not both. Entelos Inc. is a notable exception.
The company wanted to use both to support its PhysioLab technology, an in silico drug discovery and development research environment that lets companies test drug candidates on virtual patients.
“We knew there were challenges of mixing the desktops and back-end systems,” says Alex Bangs, Entelos’ chief technology officer. One issue with the desktop environment is its unpredictability. “You lose some of the deterministic nature of the system,” he says. “For instance, a user may reboot a PC in the middle of a run.”
Still, Entelos wanted the mix. “We wanted the manageability you get using back-end systems, and we also wanted to add the desktops,” says Bangs. “And we wanted a single administrative environment for desktop systems and large scale servers.”
There are open source programs that will accomplish such things, but Bangs decided to go with a commercial product. “We wanted the comfort of being able to call a company for support,” he says.
Bringing It All Together
Entelos chose Platform Computing Inc.’s LSF and LSF ActiveCluster programs to develop their distributed computing system. LSF is a workload management program that provides on-demand access to corporate computing resources.
Platform LSF supports distributed applications running on networked Unix-, Linux-, Mac-, and Windows-based servers, supercomputers, and desktop computers, according to Platform Computing officials. Platform LSF ActiveCluster is expressly for managing desktop, peer-to-peer distributed computing systems.
To take advantage of the computing resources made available using the Platform products, Entelos researchers work in the same way as if they were using a single computer. Specifically, researchers first build an appropriate virtual patient and construct medical or drug testing experiments to run against that patient.
Entelos already has PhysioLabs for diabetes, obesity, and asthma, as well as expertise in immunology, metabolism, and respiratory science. A typical simulation might be designed to study the impact of different dosage levels of an obesity treatment during several years of a patient’s life.
Once an experiment is designed, the researcher can run it locally or on a simulation server with the necessary processing power and memory. If the job is submitted to the distributed system, the researcher designates the job’s priority.
“We rate the jobs and assign them to queues,” says Bangs. “Jobs in this queue get this level of priority, jobs in that queue, lower priority.” Entelos built a workflow engine that sits on top of the Platform system.
Scheduling gives all users a chance to run their jobs. “We wanted a system where low priority jobs always get run,” says Bangs. “But only on certain processors.”
Entelos has high hopes for its distributed computing system.
PricewaterhouseCoopers estimates that in silico research such as Entelos’ PhysioLab can save drug manufacturers at least $200 million per marketed drug and cut development time by two to three years. A recognizable benefit to drug manufacturers, much of Entelos’ research is done on behalf of partners like Bristol-Myers Squibb, Johnson & Johnson, and Pfizer.
Entelos also believes the distributed system will give the company the capacity to serve even more partners. For instance, the Platform LSF and LSF ActiveCluster will support the new scientific alliance Entelos entered into with the American Diabetes Association. The research done by this joint program using Entelos’ diabetes PhysioLab technology will be shared with members of the Diabetes Research Forum.
Through the auspices of the forum, biotech and pharmaceutical member companies will have access to Compaq Computer Corp.’s Diabetes BioCluster. The BioCluster is expected to make at least a half-teraflop (one trillion floating point operations a second) of computing power and a terabyte of storage capacity available exclusively to the Diabetes Association/Entelos R&D team.