Proposal Regarding USENIX and OpenFabrics Alliance Ellie Young, 9/18/2008

For the past two months a group from the OpenFabrics Alliance and USENIX management have been working together to reach an agreement on how best to explore the feasibility and desirability of OFA and USENIX combining forces in a way that would work to the advantage of both organizations. Now the USENIX Board must decide whether we should take the next step: sending a MOU to provide services to OFA as a 501(c)(6) under our 501(c)(3). We realized last week that in order for us to announce this at the obvious juncture, the upcoming Supercomputing Conference in November, we need to have a decision from OFA by September 29.

In the past month USENIX has been given a consulting membership in OFA. Some of us have been participating in marketing, legal, and executive working group meetings. I think we have now spent enough time "getting to know one another."

Background Last year, in response to the Board's request that we reach out to open source software projects and look beyond the academic and research community we currently serve, Clem Cole introduced us to several members of the OpenFabrics Alliance. We made a proposal to organize their annual Sonoma Workshop this past spring, but they decided to continue organizing this year’s event on their own. We were contacted again this summer. At a meeting in August, part of which Matt Blaze attended, we discussed the current state of OFA. Our notes on that meeting include:

Software development: Development is the most valuable activity for OFA, but it's on its own track (Sonoma workshops, as well as workshops alongside Supercomputing, although there were none this year) and goes pretty smoothly for two software releases per year. They want to take a snapshot and make an enterprise version—one , one Windows. The desired message is that the same stack can run on both platforms. There's internal debate on the value of that, though, because and Novell need to be able to slide it into their next distribution, which poses challenges. There are 30–50+ active developers. is very active; AMD hasn't stepped up. Sun has no developers at the moment, but uses Infiniband, 10 gigabit Ethernet, too.

Membership: can now contribute software to OFA; the next step is have them become a member. With Ethernet in the software, there's "a pretty good prospect of adding members"—10 in the next six months. They need more outreach and awareness, and that could be synergistic with USENIX. Their fee structure now is probably low, and people would be willing to pay more for more outreach. Word of mouth is the biggest part of their current marketing. The Linux community doesn't link OFED to OFA, so more marketing is needed there. Mellanox has a deep footprint in this, and then QLogic. The goal is to pull in IBM, Sun, etc., to help market OF.

Training: Voltaire does their training. They would get it going with us, starting small-scale, hands-on, cluster in the corner of the room, 2–3 days long, for sysadmins, plus a "how to get management involved" no-hands-on tutorial.

Publication: We suggested that a SAGE type of booklet would be good. They agreed and said they would look around for some authors.

Working Groups: Executive, with promoter members on conference calls; annual meeting per bylaws (in theory); marketing (most active); legal (mostly concerned with licensing issues but also recently bylaws, which are being revised); enterprise working group for Linux (in the enterprise group, there's a release coordinator); Windows; interoperability (with plugfest event 2x a year in NH, if you pass you get a logo—they pay $20k/year to Lamprey to maintain and improve the tests); user working group supposedly does outreach to end users to provide feedback and guidance, but has been fairly inactive.

The Future of OFA and Other Ideas for USENIX: RDMA technology is in its infancy now, so it's got a long future. Clem says it's a work in progress. Bill Boas thinks OFA is logically but not physically a company, creating two products. They need a process for developers to learn what users need, a calendar, project planning and management. A process for improvement is not now in place. They need regular interaction with users. If USENIX is interested in supporting the creation of OS software, we can get a process from OFA. We might be able to reach out to Hadoop. OS filesystems is also a wide-open community. Perhaps we could lure Fibre Channel into becoming open instead of proprietary. Management of clusters is another area needing open unification.

What OFA Hopes to Gain

Their volunteer model isn't scaling, and they would benefit from USENIX providing infrastructure. At this time that would include:

• Organizing and marketing their annual Sonoma Workshop • Marketing push for new members • Staffing at their booth at Supercomputing and promoting membership in OFA there • Promoting OF software at USENIX events (esp. LISA and FAST) and offering training at USENIX events • Providing financial and administrative services • Establishing a training program for OF software • Monitoring/administration of all OFA working group meetings, including IWG • Forming and providing momentum for an improved developer community • Hosting and maintaining the content on the OFA Web site and hosting the software repository Web site • Managing the contributions, provenance, testing, and release of OF software • Making the Sonoma and Developer workshops part of the USENIX community

What USENIX Hopes to Gain

Clem wrote email to some of the OFA members on what USENIX's goals are in bringing OFA into USENIX. See http://www.usenix.org/bodinfo/bod/sept08telecon/clem_ofa.pdf.

Potentially there’s a market for new corporate members, including labs, for USENIX if we join with OFA.

Other OSS projects/developer groups may reach out to us (e.g., Hadoop). OFA has created a process that can be adopted by other groups. Is there anything in the cluster/grid/cloud arena that needs cooperation because no one is in a position to own it (loaders, system management, resource management, protocols, interconnect, security, etc.)?

Potential Problems/Hurdles

OFA governance/structure: They currently have a large Board, all volunteers, and everyone gets a vote: this is not a scalable model. Also, we have no idea if, once USENIX is on board, the volunteers doing most of the work in development, licensing, content for conferences, etc., will disappear. They may see USENIX as a management services company (which, of course, we aren't). We need to have them make the leap to becoming an OSS project team.

Staffing: Ramp-up with USENIX staff will initially be steep, because we’re dealing with a different model/community.

Code development: USENIX has never openly shepherded this type of activity. Would this make us more like OSF/The Open Group? How would this mesh with the rest of our program/offerings?

Conferences: USENIX has made its reputation over the years by its highly technical, peer-reviewed conference proceedings. The Sonoma Workshop is hardly at that level. How might this affect our reputation? See programs at: http://openfabrics.org/archives/april2008sonoma.htm http://openfabrics.org/archives/april2007sonoma.htm

Marketing: We could be viewed as weak in marketing, since we have no experience in product marketing; our efforts have been focused on event marketing. We are also not strong in the area of industrial relations. However, OFA may be a vehicle for us to get more experience in these areas.

Finances: OFA has supplied us with all the information we have requested. They have assets between $150K and $200K (they call it their reserve fund; its purpose is mostly to cover them in case their conference attendance is unexpectedly low). In 2007 they had a overall net revenue of $100K, and they estimate a $60K net this year. There are no liabilities and only one signed hotel contract for 2008. I have the budget and tax returns, if anyone wishes to see them.

Legal issues: I checked with Dan Appelman on the best course of action for setting this up. I particularly asked about the effect, if any, such a move would have on our 501(c)(3) status. To bring this group into USENIX as a SIG by definition would mean they would cease to exist as a stand-alone legal entity and would just become part of USENIX, relying on USENIX's tax exemption. They would probably sell their assets to USENIX via an asset purchase agreement and then wind up and dissolve the corporation.

While this is the simplest way to proceed, we do have a concern on whether the mission of the OFA would qualify for 501(c)(3) status if it were a stand-alone entity. The basic requirements are that it either be educational (which USENIX clearly is), scientific, or cultural, and that it can't be promoting some commercial standard or direction for technology. While I think we would have a good chance of eventually making the case that OFA as a SIG would qualify, I think it may be more prudent to have it remain a 501(c)(6). USENIX can provide services to a (c)(6) as long as we properly allocate, keep some accounts separate, etc. I advise that we suggest that OFA remain a (c)(6) for the next two years and that we plan to revisit this at that time.

Should the Board approve our going forward, I will work with Appelman on a MOU that will include the scope of USENIX's services, the compensation USENIX would receive, and who has the ultimate authority for making decisions.