When CERN’s Large Hadron Collider smashes particles together at near to the speed of light, it creates a spray of smaller particles that hint at the nature of the universe. It also creates a lot of data that must be accessible to physicists all over the world.
Tim Bell, who will speak at GigaOM’s Structure Europe in London Sept. 18-19, is CERN’s infrastructure manager. Bell started out as a Unix kernel developer at IBM before moving to Deutsche Bank to manage its infrastructure. He has now been at CERN for eight years. Every day, his team makes sure operating systems, mail and security infrastructure are functional to serve 11,000 physicists. They increasingly also oversee cloud computing resources that make working with huge amounts of data more manageable.
“It has become a key tool for physicists to able to understand the nature of matter and the universe itself,” Bell said. “The ability to turn around rapid amounts of analysis and allow the physicists to explore the results [is part of] delivering discoveries.”
CERN maintains 11,000 servers at its campus outside Geneva, Switzerland, that build preliminary reconstructions of particle collisions before sending data to smaller centers all over the world. Roughly 600 million collisions a second are whittled down to 60,000 to make them possible to compute, and then 600 are selected as especially interesting for further review. That results in about 25 petabytes of data that need to be stored each year.
Last year, CERN added a second primary data center in Budapest that now has about 1,000 servers, but they did not have the resources to add additional staff. Bell said they looked to see how other organizations were running large data centers and cloud computing emerged as an option. Their data centers’ resources are now managed in OpenStack–an open source program that physicists can use to request resources like memory and storage space. Bell serves on OpenStack’s board of directors.
“It allows the physicists to come up to a web interface and obtain a new machine in the order of 2 to 10 minutes,” Bell said. “Previously when they asked for physical software they would be waiting for months in comparison. You have an idea, you ask for resources, you go away, get a coffee and when you get back those resources are available for you. “