Original URL: http://www.theregister.co.uk/2012/07/16/intel_buys_whamcloud/

Intel gobbles Lustre file system expert Whamcloud

Chipzilla prepares to gorge itself on exascale pie

By Timothy Prickett Morgan

Posted in HPC, 16th July 2012 11:43 GMT

Intel's ambition to dominate the exascale era of computing is a little closer to realisation, as Chipzilla has written a cheque to acquire Whamcloud, a company which develops and supports the open source Lustre clustered file system.

Brent Gorda, president and CEO at Whamcloud, quietly announced the deal late last Friday.

The chip giant already basically owns the x86 server processor market and has taken big strides into the market for CPUs in embedded devices and storage arrays. Networking is another market in which Chipzilla shows signs of possible dominance, thanks to relationships with Fulcrum Microsystems, QLogic and Cray.

Whamcloud gives Intel a strong foothold in exascale computing, as the company recently took down a contract with the US Department of Energy for an exascale search system. Intel later announced it had its own contracts under the FastForward section of the Extreme-Scale Computing Research and Development program, the primary funding vehicle for the US government's exascale efforts.

"The Whamcloud acquisition extends Intel’s software and services portfolio in the high performance computing space in addition to reinforcing Intel's position in the open source community," Gorda wrote in a statement posted to the Whamcloud web site. "Working as one company, we are now in a stronger position to advance our mutual goals and continue providing vendor neutral solutions, delivering greater value to our customers, and moving the industry to exascale performance."

The financial terms of the Whamcloud acquisition were not provided. Customers are being instructed to interface with their Whamcloud sales and tech support teams as usual. Gorda has been named general manager of the Higher Performance Data division at Intel.

The Lustre project was founded by Peter Braam, a researcher at Carnegie Mellon University who in 2001 created his own commercializer for the code, called Cluster File Systems. In September 2007, with its eye on the HPC market, Sun Microsystems bought CFS and intended to merge some of its Zettabyte File System (ZFS) capabilities into Lustre. Oracle bought Sun in January 2010 and technically controls the open source Lustre project. But the company has other file systems it likes better and those are not focused on HPC - BTRFS and OCFS2 are the key ones - but rather on more traditional databases and applications.

Whamcloud was established in July 2010 with the express goal of extending and supporting the Lustre file system, which is used in more than half of the supercomputers listed on league tables of the globe's 500 mightiest machines.

Whamcloud got $10m in private funding to form and hired Gorda, who use to work for the DOE and who actually cut the checks for Lustre to be developed at Lawrence Livermore National Laboratory, to run the company. Eric Barton, who was CTO at Whamcloud, was the technical lead for Lustre development at LLNL and then at Sun, is a key asset for anyone doing Lustre development. So is Robert Read, who was in charge of Lustre 2.0 at Oracle and who joined Whamcloud as principle engineer when it was two and a half years ago.

The latest development from Whamcloud on the Lustre front is a management tool for the parallel file system called Chroma Enterprise, which came out of beta last month at the International Super Computing conference in Hamburg, Germany. Lustre may be scalable, but it sure is not known for being easy to manage, and that is one of the things that Whamcloud – and now Intel – will be working on some more.

Whamcloud's acceptance of Intel's embrace could also mean Chipzilla does something really funny, like get back into the main memory business, which it exited in 1985. Memory is a bottleneck in exascale systems – perhaps the bottleneck – and if memory is going to be packaged onto processors in 3D chip complexes, then Intel has little choice but to either partner strongly or do memory itself. Intel really prefers to do the engineering itself when possible because that is how the company maintains control, builds patent portfolios, and wrings out profits. ®