Lab organizations looking to buy high-performance computing systems, such as cluster computers, will find their procurement request being handled by the Computing Sciences Acquisition Team sooner or later. And the team, which has experience buying systems ranging from small clusters to some of the world’s biggest supercomputers, would like to get involved in the process sooner, rather than later.

“It’s better for everyone involved if the project comes to us early on,” said Gary Mack, a senior subcontracting administrator on the team. “We can help you avoid those ‘gotchas’ that can delay the project.”

Best of all, there is no additional charge for enlisting the team’s assistance early in the procurement – they are paid for by the procurement burden, which is based on the total amount of the procurement, not the time spent by the team on the purchase.

Suzanne Nolan, head of the team, recommends that Lab staff members interested in buying multi-processor systems contact the team before talking to vendors. Sometimes even off-hand comments to vendors have led to confusion, difficult negotiations and even formal challenges.

Meeting with the procurement team to map out system requirements and expectations, and clearly identify budget constraints is the best way to start. The team can then solicit bids for a system to meet those requirements.

There are also different ways to solicit bids. In a competitive bidding situation, the Lab advertises that it is looking for a well-defined system with very specific processors and software.

A second approach is called “best value” and typically includes a short list of requirements and a longer list of desirable features. Vendors are then asked to submit proposals on what they feel is the best solution, which may not have been immediately apparent. Such an approach was used to procure the NERSC Center’s 6,656-processor IBM supercomputer, as well as smaller systems used by researchers in several Lab divisions.

As part of the contract, the procurement team can also include acceptance criteria, which state how the system is expected to perform in a demanding scientific production environment. If the system doesn’t perform as expected, the team can go back and negotiate for a better solution. “Including this criteria in the contract gives us some leverage to ensure that the computer meets our needs,” Nolan said.

Mack added that the team has now developed a Request for Proposals (RfP) for buying clusters and that this can be revised slightly for each procurement. Having such an RfP on the shelf can also get the procurement process moving sooner.

“We really want people to call us even when they’re just thinking about buying a system so we can work with them each step of the way,” Nolan said. “That’s what we’re here for.”

Contact the Computing Sciences Acquisition Team (GAMack@lbl.gov).