Grid running strong

29 September 2008

A Google image of the Grid infrastructure in Europe, with a few points visible in Asia and North America

 

The accelerator will have its first collisions in a few months, and the detector is ready to collect data. But will that data, when it comes, have somewhere to go? In other words: Is the Grid ready?

If you ask Dario Barberis, computing coordinator for the ATLAS experiment, they’ve been ready. Well, those weren’t his exact words, although he affirmed: “We’ve passed the point of testing.”

The folks on the Worldwide LHC Computing Grid made sure that they would beat the LHC and detectors to completion, and as a result, they’ve been prepared for LHC data since early summer. After all, there’s no point in having a working experiment if you can’t record the data.

“The Grid has been in run mode for over five years, for simulation production and analysis,” says Dario. Since early summer last year, the Grid has been sending cosmic ray data as well to ATLAS’s ten Tier 1 computing centres and 50 or so Tier 2 centres.

“We distribute data continuously,” says Dario. They’ve been pushing the limit of the system, exporting up to 1.2 gigabytes per second out of CERN, or one DVD every four seconds.

The Grid is also handling analysis jobs – somewhere around 60,000 per day at the moment, but they expect between 100,000 and a million per day once the LHC is colliding protons. Analysis on the cosmic data and the early beam data is used to tune ATLAS. According to Dario: “People use these data to exercise the calibration, alignment, and reconstruction procedures.”

While each computing centre has its own particular set-up, the Grid middleware evens out the software landscape and unites them so that each collaboration’s computing assets can be accessed as though they are a single, giant computer. Above the middleware, ATLAS has its own software that allows ATLAS collaborators to interface with the system and run jobs.

The average job takes anywhere between three and 12 hours, but heavy ion physics can require an entire day to run a job. “We try to limit them to less than one day per job because it’s a more efficient use of the resources,” Dario explains.

This winter, while the LHC is still shut down, the Grid centres are already looking ahead to a substantial upgrade of several million Swiss francs in total. The cost will be distributed among all the data centres and these upgrades have already been built into the centres’ budgets. “For each one, it’s not an enormous amount of money,” says Dario, and it will double their capacity, in terms of both storage and processing.

 

Katie McAlpine

ATLAS e-News