I learned something interesting about the new SGI super being fired up in New Mexico: the state intends for it to “pay for itself.”
Sue Vorenberg has a profile piece on the machine (named Encanto, Spanish for “enchanted”) in the Santa Fe New Mexican (here) that focuses primarily on the visualization challenge — getting from data to information, as we say in the trade.
[Update: Gerardo Cisnero from SGI wrote to tell me that Encanto is actually Spanish for "enchantment", a reference to the New Mexico state motto: "Land of Enchantment." You would think that the New Mexican wouldn't have missed that one.]
I haven’t been paying too much attention to how the business of this machine is getting done, so I found it interesting to learn that the state intends for the machine to support itself by charging for cycles
The Computing Applications Center also is hoping that the Research Applications Act, House Bill 262, sponsored by Rep. John A. Heaton, D-Carlsbad, will pass this session. The bill, which is in the House Business and Industry Committee, would set up how the facility will operate and switch its management to a 501(c)3 nonprofit organization that could charge companies for use of the system.
That is how the project will become self-sufficient at the end of five years, said [Lorie Liebrock], who also is the center’s interim education director.
I do think that a hosted model for HPC cycles will arise to serve those who either aren’t using HPC today, or who are on the fringe with small systems they’d rather turn over to someone else. But I’m thinking right now that those businesses are likely to rise organically out of large ISPs and co-location centers operating today who already know how to deploy large scale resources used in small chunks by many customers.
A resource on the scale of the New Mexico center is built for large scale codes, which right now means big science. Big science doesn’t (usually) have a bottom line motive, or a budget for computing cycles, but does currently have a free source for cycles (depending on who’s paying for your science, DoD Mod, the DOE, and the NSF centers all come to mind). So they’re not coming to the table with money.
I suppose that NM universities could be taxed or otherwise forced to allocate part of their budget to compute on that machine, in part by a ban on deploying clusters and HPC machines outside of the new HPC center. But this type of “managed economy” is only sustainable under a strong regime with a devotion to HPC — put in a new governor, or change some part of the policy to respond to a more pressing political issue (like, say, a tanking economy), and HPC is suddenly not so important.
So it seems that revenue from chargeback would have to come from people with a budget — businesses not currently using HPC. This means small businesses, engineering consulting firms, and the like. This kind of user needs an infrastructure built to support their transition from workstations to the extremely user hostile environment of supercomputing. From what I can tell, the state has put almost all of its money in building the machine and its satellite access centers. In fact, it is still working the legislature for operations money
The New Mexico Computing Applications Center, which is the overarching title for the effort, also is trying to get a $5.8 million special appropriation for operations and equipment from the Legislature to add more gateways, fund staff and some research projects, and cover the annual operating costs of $1.9 million, [Thomas Bowles, Gov. Richardson's science adviser] said.
There are no real commercial products out there right now that a center can deploy to make HPC user friendly (although there are important initiatives in this direction for example at OSC, in the DoD Mod Program, and through the Council on Competitiveness), which means that in order to actually attract the HPC customers it needs to make chargeback work the NM center is going to have to invest some money to bridge this gap itself. Money it doesn’t currently have.
Richardson said he foresees the system fostering statewide water modeling projects, forest fire simulations, city planning and the development of new products, and as a lure to bring more high-tech industry to New Mexico. “It is not simply a high-tech toy for elite scientists,” the governor said. “This project invests in our future.”
I think the New Mexico investment is a Very Good Thing. But the state needs to abandon its hope of this particular machine paying for itself, and use it instead to attract new research-based initiatives to New Mexico. Smart people are attracted by great tools. Make this one available to them, and the governor’s dream of using supercomputing to kick start the state’s high tech development could still come true.