At the OpenStack Summit in Paris, France, CERN’s Infrastructure Services Manager Tim Bell gave the general session audience an overview of his institution’s experiences moving to OpenStack, which he characterizes as a “cultural and technology transformation.”
CERN, the European Organization for Nuclear Research, supports 11,000 physicists from around the world. These scientists use the facilities to conduct basic research in their quest to understand what the universe is made of and how it works.
CERN was behind the famed Higgs boson confirmation in 2012, but the Higgs wasn’t the only fundamental question sought by CERN scientists. Physicists remain puzzled about the nature of matter and antimatter. “When we count the planets and the stars, for example, we see that we’ve only got 5 percent,” says Bell. There is something out there – theorized as dark matter or dark energy – which must be present to explain why the cosmos behaves as it does, he adds.
Another fundamental question concerns gravity. Scientists can describe three of the four forces very well with the standard model, which the Higgs helped confirm, but gravity is a real problem. Physicists theorize that there are particles called gravitons that move in and out of other dimensions.
“As we move the LHC further on, we hope to discover some of these particles and understand the universe further,” says Bell.
Solving problems of this magnitude requires a large dedicated community and well-constructed experiments. Conceived in the 1980s, the LHC consists of a 27-kilometer ring 100 meters underground on the Franco-Swiss border. It was designed to collide beams of particles just below the speed of light.
Detectors observe and record the results of these collisions, taking 40 million pictures a second.
“That creates, amongst other things, some great pictures,” says Bell, “It also creates one petabyte per second of data.”
To handle this massive data stream, CERN has relied on very large computer farms, also 100 meters underground, that filter the data to levels they can record for further analysis.
Still, the experiments around the ring generate up to 27PB of data each year, which is expected to be saved for 20 years. By 2014, CERN had amassed a 100PB archive, primarily stored on tape. In April 2015, the accelerator will come back online after an upgrade to double the energy of the beams. This will result in even higher data rates.
But CERN is looking further ahead. By 2023, they anticipate an annual data load of 400PB, requiring a 50-fold increase in compute power.
CERN needed an environment that would scale to handle these massive needs. Their main Geneva datacenter was equipped with one mainframe and one Cray. Using standard industry servers, they cannot fill up the empty racks that line the datacenter without going over the 6kw per square meter – the max that this environment can cool.
To expand capability, CERN established an additional datacenter in Budapest, which is now online, linked to Geneva by dual 100GbE connections. Unfortunately the current economic and political reality is such that: staff numbers are fixed; the materials budget is decreasing; and legacy tools are high maintenance and brittle. Despite the limitations, users expect fast self-service.
CERN’s primary challenge then was to bolster IT services without increasing support staff. This prompted CERN to investigate new infrastructure tools and processes. They deduced that from a computing point of view, there is no reason to be special. Regarding the staffing situation, there is no Moore’s law for people; therefor automation needs APIs, not documented procedures. Culturally, they looked to open source communities and models for inspiration.
After much discussion, research and prototyping, CERN selected OpenStack to bring a flexible and agile cloud to their users.
They started with what was essentially a research project in 2011 with Cactus. Immediately, says Bell, it was clear that the rate of maturity of the software was going to exceed the rate that CERN would be reaching production on its own. After a period of training and tooling, they went into production with the Grizzly release in July 2013.
Currently, CERN operates four OpenStack Icehouse clouds. The largest is currently around 75,000 cores on more than 3,000 servers. There are three other instances with 45,000 cores total located at CERN’s underground compute facility to deliver additional simulation capacity. They have another 2,000 additional servers on order, and will be passing 150,000 cores in total by the first quarter of 2015. All the code that is of any interest to the community has been submitted upstream, and all CERN-specific code is publicly-available on github.
OpenStack’s Nova Cells feature will enable CERN to scale to meet its needs in the near-term and in the future. The Cells approach lets them build up small units of OpenStack that can be assembled together to appear as a single homogeneous resource. It simplifies the end user experience while still scaling out the underlying environment, says Bell.
CERN was also able to address the problem of working across multiple clouds. With help from Rackspace, CERN developed federated identity capability on OpenStack, and the code for this is now in production release.
“So remember,” Bell tells the audience, “whenever you’re helping out OpenStack, you’re helping us understand how the universe works and what it’s made of.”
Earlier at the OpenStack Summit, CERN was announced as the first winner of the OpenStack Superuser Awards in recognition of their accomplishments and community involvement.