Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
August 1, 2011

The State of the Lustre Community

Brent Gorda

A year ago the Lustre community was stunned by Oracle’s message at the 2010 Lustre User Group (LUG). Lustre was no longer a vendor neutral platform; you had to buy Sun/Oracle storage hardware to get future versions of the software. The community uproar was strong to the threat HPC’s most popular file system going away. As a result, a flurry of activity ensued with the formation of multiple community groups and startups, including the founding of the Lustre-focused startup Whamcloud.

The Lustre community quickly organized meetings in Europe at the Jülich Supercomputing Centre to self organize and react to the issue. Because the Lustre code is under GPL (GNU General Public License), the threat of a fork, or multiple forks was in the air and everyone wondered what would be the fate of the technology.

Out of the meetings, not-for-profit community groups were formed including:

  • EOFS — The European Open File System Cooperative (Europe-based)
  • OpenSFS –The Open Scalable File System (US-based)
  • HPCFS — The High Performance Cluster File System (US-based)

Everyone had the best interests of Lustre in mind and opinions on how to best care for this critical bit of technology that so many were dependent upon.

The two US-based groups initially differed in their approach, a concern for those not paying close attention to the players and their goals. OpenSFS raised a significant amount of funds and pledged to lead the community through continued investment by the DOE labs (LLNL, ORNL) and interested HPC vendors (DataDirect Networks, Cray), the original members of the organization. Continued support from these large players would, OpenSFS stated, ensure the longevity of the technology and its availability to all. The argument made good sense as that was the environment in which Lustre was born and matured.

The HPCFS approach was to organize the community and leverage the resources that exist at member sites. By having minimal fees, the membership would thrive and self-organize to do the work necessary to preserve and move the technology forward. This focus on grassroots strengths saw that the large outpouring of interest in helping Lustre would carry the weight of caring for the source code, making releases and providing support.

In Europe, the community debated the issues and formed EOFS after the well-planned and attended meetings in Jülich. The organization took time to ensure that all voices were heard and in December of 2010 the founders gathered in Munich to officially sign the documents necessary for a non-profit cooperative to be headquartered in any of the European Union participating countries.

For several months leading up to the 2011 Lustre User Group, the three groups communicated to Lustre sites, within their organizations and across organizations. The feeling was that there was a bit of an overreaction and, as time passed and people communicated, everyone began to realize that we are all on the same team with the same goal of preserving the Lustre technology for the community worldwide.

At this year’s 2011 Lustre User Group, the two US-based groups announced a merger that was completed a few weeks later. At the International Supercomputing (ISC) show in Germany, shortly thereafter, the two existing community groups, OpenSFS and EOFS, signed a memorandum of understanding to show their allegiance.

Naysayers had been predicting a fork in the code and a fracturing of the community. Alternate solution providers were putting the scare on — spreading fear and doubt as to the future of the technology. In the end, however, they have all been proven wrong. There has been no fork, there is one base tree from which everyone has agreed to make their releases. And the community has not fractured. Instead, it has pulled more closely together than ever.

It’s important to mention the major companies that have been responsible for developing and spreading Lustre. In the past few years, the technology has seen quite a bit of change in corporate stewardship. In the early days, it had been closely held in a small company Cluster File Systems. Sun pledged support for HPC and acquired CFS but their big push into the HPC market faded with the economy. Finally at Oracle, community concern was based on a perceived lack of interest in anything HPC.

I think that we in the community owe these companies credit for the excellent state that Lustre is in and the fact that it is still fully available as open source. Lustre is, today, a mature and stable technology due in large part to the significant investment made in the Lustre engineering teams over the past several years. As just one example, through the Hyperion consortium at LLNL, Sun and Oracle invested in equipment and, significantly, in the engineering time necessary to mature the technology. This should not be overlooked.

Today the HPC community finds itself in excellent position. Through the efforts of all the groups, the community at large and the steadfast confidence of the engineering teams, Lustre has emerged in fine form and stronger than ever. As we look forward to embarking on the long path to exascale, Lustre is an obvious technology choice for the journey. At the end of this estimated 8 to 10 year effort, we may not recognize the code –it may all be replaced by then — and the name may be different, but the platform we are starting with is second to none in the industry.

—–

About the author

Brent Gorda, Whamcloud CEO and President, joined Whamcloud from the US Department of Energy where he was involved in program funding and strategic adoption of the Lustre File System at the Lawrence Livermore National Laboratory and other ASCI labs.