A very insightful article about open source business models was posted by Mike Olson, Chief Strategy Officer at Cloudera. By way of background Mike Olson was involved in Michael Stonebraker's Illustra data base startup, which I remember as the first database that extended DBMS support to spatial data types. Michael Stonebraker called this type of DBMS, object-relational. I still have Stonebraker's book called Object-relational DBMS, The Next Great Wave published in 1996. The basis for the project was Postgres developed as an open source project at UC Berkeley under the BSD license.
According to Mike Olson, for Illustra Stonebraker relied on the approach that he had made successful with Ingres in the 1980s, which I also remember. He created a proprietary fork of Postgres (now known as PostgreSQL) and made Illustra a traditional, closed-source software company to sell it.
In 1998, Mike Olson joined Sleepycat, who made and open source embedded database called Berkeley DB, now owned by Oracle. Sleepycat made money by "dual licensing" BerkeleyDB. Dual licensing was quite widely used as I remember. You could either use software under a GPL licence, which meant that if you added any of your own code, you had to make it available under the GPL. But there was a way that you could keep your own code proprietary, by buying a license from Sleepycat. Apparently this was pretty lucrative for Sleepycat. I think MySQL (now part of Oracle) and others had a similar type of dual licensing.
Another open source business model that a number of folks used was selling services and support - basically programmers' time to modify or enhance an open source project. Organizations would pay open source developers to add a new file type or new functionality to an open source project.
Mike Olson makes the case that there were very few successful large stand-alone open source vendors. He counts Red Hat as the only one that has not been acquired or has not failed. But he says that Red Hat did not become successul purely through a support and services model, but through its closed-source Red Hat Network product, which it made open source after eliminating the serious competition.
The problem Mike Olson sees with a services and support model for open source is that as soon as a vendor becomes successful, meaning getting large enough to be noticed, it gets the attention of good, but low cost programmers in China, India, and Eastern Europe, who have access to the source code. Ultimately this drives down the price yu can charge for support and services.
The other trend that Mike Olson points to is ubiquitous open source software infrastructure including operating systems (Linux), web servers (Apache), scripting language (Python, Perl, PHP), databases (MySQL), enterprise programming language (Java), virtualization (Xen), and so on. Closed source is a diminishing part of this business.
So the quandary for folks who want to be in the enterprise infrastructure software business is, how do you create a large successful, stand-alone software company when the platform has to be open source, but pure play open source software companies historically haven't survived when they get large.
According to Mike Olson, the model that Cloudera has found to be successful is to build value for custmers as closed-source, proprietary IP that differentiates Cloudera on top of an open source platform Hadoop. Clearly this requires commercially oriented, non-GPL open source licensing for the platform, such as the Apache or Eclipse licenses.
Thanks to Andrew Ross for pointing me to this.