Clustering is moving forward in the BCRC. One of the Darwin Fellows had approached me during the semester about using MrBayes, a package for doing phylogenetic inference using Bayesian estimation. I tried to find time during the semester to work on it, but was simply too busy to get it set up. But now, there's time! I had hoped that Xgrid might be the ticket, but it's not really the right thing. MrBayes uses MPI, but xgrid doesn't do MPI. Luckily, MPI compiles, installs, and runs cleanly under OSX, so it's simply a matter of getting it integrated into the lab's radmind image. We're about half-way there. One nasty aspect of OSX is that HFS+ is case-preserving, but not case sensitive. MPI produces two scripts for compiling C and C++ code: mpicc and mpiCC. Unfortunately, with HFS+, you simply can't have both! Argh! You can get around it, by building it on a UFS formatted filesystem, but what a huge pain. I always knew Apple's decision to use HFS+ instead of a real file system was going to come back to bite us over and over again. Oh, well.


Lately, I had been feeling a bit disappointed with the student buy-in on Muppyville. There were a few students who showed extraordinary interest, but one of them I had to ban for causing problems and I feared the others didn't form enough of a core-group to maintain interest. But this morning, I checked the lastlog and discovered that 8 students logged in last night to chat with a classmate who's gone back to Bolivia for a year! I'm so pleased! That is exactly the sort of thing I'd been hoping students could do. I've been meaning to get a login for another classmate who moved back to Egypt last year -- I'll bet they would have fun chatting with him too. In any case, I'm feeling quite a bit more hopeful than I was.


We got MrBayes working! It was a huge all-day ordeal, but it works! We spent the morning getting MPI working and then in the afternoon we were stumped when the copy of MrBayes that we had compiled with the MPI libraries didn't seem to be working in parallel. Once we looked at the source, it was obvious that there was another define that needed to be set. The rest was just putting the icing on the cake: building a directory that would automount for the data that needs to be available on all the machines and tweaking the script that runs radmind so that I can easily set to not run -- otherwise the machines will all reboot in the wee hours of the morning. And now we can start looking at more new and exciting ways to use MPI!


StevenBrewer