I have managed to do it again. I’ve overloaded myself w/ too much to do, with too little time. Thankfully my biggest near-term concern is nearly complete. My OLS paper is almost finished. The paper is on the genetic-library in the Linux kernel. It will be very useful for people to learn about the genetic-library and hopefully show how beneficial it is.

The biggest problem I have encountered is a regressions in performance results on the Zaphod CPU plugin. I was getting only 1%. Thankfully, my co-author, Peter Williams, was able to track down a bug in the run-delay calculation, and get the results back up to the >3% that I was originally seeing. I also noticed a trend towards the Zaphod plugin doing very well in higher CPU loads.

The best news I have gotten from doing this paper is stellar results in the Anticipatory I/O scheduler. On average, I was seeing 8.72% improvement, and up to 23.22% improvement in the random write workload.

The next thing I need to look at is a combination of RAID’d and unRAID’d disks. I was seeing huge fluxations in results on Anticipatory (which should not be used on RAID’d disks in the first place). I am theorizing that because the genetic-lib is per I/O scheduler, that it bounces back and forth for tuning for the individual SCSI disk, and then tries tuning for the RAID’d disks. Which one is tunes for depends on which one is getting more I/O. I will have to granularize the genetic-lib to operate on a per-disk basis. Thankfully there has been some support put into the kernel recently.

Watch http://linuxsymposium.org for updates.

I will hopefully have some time to do some more development work on the genetic-library soon, and a new version coming out soon.