August 30, 2010

LHC Compute Grid Teaches Some Valuble Lessons

Nicole Hemsoth

CERN’s Large Hadron Collider has proved to be an interesting test case for ultra-large-scale data collection and computation. The project, which generates petabytes of data for its particle acceleration mission, requires one of the largest and most widespread compute grids in the world. Less than a year after the collider began fully operating, the LHC project is already blazing a path for eScience.

The dominant computational theme of the LHC work is data reduction. With petabytes of data being generated every second by the LHC detectors, the challenge is to filter out the uninteresting information so that the critical data can be more easily sent to secondary sites for storage and processing.

According to an Ars Technica article penned by John Timmer, in general, the LHC cyberinfrastructure is performing even better than expected. Although a 35-year-old datacenter is forcing higher density compute clusters (and water cooling), the robust network and improving price-performance of disks have lessened the project’s reliance on tape storage. Writes Timmer:

One of the reasons for the increased reliance on disks is the network that connects the global grid to CERN. “Because the networking is going so well, filling the pipes can outrun tapes,” von Rueden told Ars. Right now, that network is operating at 10 times its planned capacity, with 11 dedicated connections operating at 10Gbps, and another two held in reserve. Each connection goes to one of a series of what are called Tier 1 sites, where the data is replicated and distributed to Tier 2 sites for analysis. Von Rueden said that the fiber that powers this setup has been “faster, cheaper, and more reliable than in planning.”

An interesting aspect to the LHC setup at CERN is that they’ve decided to limit hardware support contracts to no longer than three years. The rationale is that because price-performance for hardware is rising so rapidly, it doesn’t pay to keep any particular machine running too long; when it breaks down, just replace it with something cheaper and faster.

Full story at Ars Technica