A look at the Bizgres Project

The Bizgres project was launched in April of this year. It is based on the PostgreSQL project, with development sponsored by Greenplum, which also uses the Bizgres source code in its DeepGreen offering. Josh Berkus, who works for Greenplum and is a member of the PostgreSQL core team, talked to us about the Bizgres release and the plans for the project.

The 0.6 release has two features of interest that are not currently found in PostgreSQL. The first is a patch that speeds up bulk loading of text data. Berkus said that the patch "speeds up bulk loading of text data by refactoring some of the bulk loader code."

The other feature is an improvement in temporary table creation. When tables are created using the "CREATE TABLE AS SELECT" statement, transaction logging is bypassed which can provide major benefits in performance -- in effect, a kind of "scratch" table that can be used to hold a copy of data that is being worked with without logging transactions. Berkus said that the Bizgres team is planning on expanding the capability to include the ability to bulk load into a "scratch table" but the current feature does not allow that.

With so many PostgreSQL-derived projects available, some may wonder if the project is forking. He said that Greenplum plans to contribute its features back to the PostgreSQL project, but that the timing of PostgreSQL releases made it hard to get the features that interest Greenplum and its customers into mainstream PostgreSQL in a timely fashion:

Since PostgreSQL 8.2 is currently slated for summer of 2006, there is a distinct advantage in creating a derivative distribution of PostgreSQL to allow Greenplum and the Bizgres Project to push its features out to its users earlier. Berkus compared this to Linux vendors like Red Hat backporting features from the 2.5/2.6 kernel series to the 2.4 series while 2.6 was still in development. He also compared Bizgres to embedded Linux or Real Time Linux, "in that we're focusing on a distribution of PostgreSQL entirely focused on needs of people running data warehouses or doing business intelligence." Users outside those profiles, said Berkus, probably don't want to consider Bizgres or DeepGreen at all.

He also said that the Bizgres project is focused only on Linux, Solaris and Windows, as opposed to all of the platforms that are supported by the PostgreSQL project, which produces fewer platform compatibility issues for Bizgres.

Berkus allowed for the possibility that Bizgres could have features that do not make it into mainstream PostgreSQL, if they were of benefit to data warehouse applications without providing a benefit to general performance, but that he wasn't aware of any features under consideration that would fit that category.

As for licensing, Berkus said that anything developed by Greenplum for Bizgres would be available under a "BSD or analogous license."

He also said that he wasn't concerned about other companies snapping up Bizgres' technology. According to Berkus, the major vendors like Microsoft, IBM and Oracle, "already have technology of their own that they have investment in, and they're unlikely to abandon theirs... and if they did [take Bizgres features and make them proprietary] it would be enough of a moral victory that it would be worth it."

Given the number of companies working on PostgreSQL distributions, it should be interesting to see how many of the improvements flow back into the main project, and whether the various companies can avoid straying too far from the main project. It should also be interesting to see whether the Bizgres project gains much steam as an independent project. The mailing list traffic isn't particularly heavy yet, but the project is still very new.

For users who are interested in trying out Bizgres, the 0.6 release is available as source code or binaries for Solaris 10 or Red Hat Linux.