Ummm ... pause for stretching the ol' credulity muscles. A decrease from 17 hours to 11 minutes implies somthing quite a bit more than a switch from some labeled Unix to Linux happend. <EOM>
bump.
Perhaps those ols Sun minis aren't really as fast as claimed. DB performance is heavily influenced by how many indexes can be held in memory. You can get 3 gigs of DDR on a cheap motherboard in the PC world.
They probably went from having (32) Sun SPARCStation 10's (SPARC CPU @ 20Mhz) running Solaris to (40) Intel boxes (2.4 Ghz Xeons) running Linux.
They probably couldn't justify the cost to replace all 32 Unix boxes with their modern equivilents.
If the Unix boxes were old enough to be ready for replacement, the new hardware is probably 1 or 2 orders of magnitude faster. They should have compared running Unix vs. Linus on the same hardware, to make a comparison that means anything. I'm sure that's where most of the performance increase came from.
Really tight resources, and the OS clogging up those few open resources could do it. We've got a system that just a slight reduction in inputs makes the whole thing run a lot faster. Kinda like contention problems.
Actually it doesn't surprise me at all. While it is true that Linux (and FreeBSD) tend to be substantially faster for compute intensive functions on the same hardware compared to Windows or commercial Unix variants, the specific instance in question was probably compared to some over-priced and somewhat crusty Sun boxes, which have never delivered stunning performance even when new. I actually have seen this kind of performance improvement when upgrading systems from a commercial Unix to a free x86 variant. If you actually analyze it carefully, there is nothing particularly fantastic or stunning about it because the x86 hardware is frequently scads faster.