Monday, May 09, 2005

Adventures in Frustration

On the one hand, I am so close to graduating! On the other hand, I'm so close to graduating! Aargh!

All my runs are completed. Now I'm just trying to make sense of the data. In particular, I am having trouble fitting a sensible model to my data. You see, my program has the curse of superlinear speedup.

Superlinear speedup is actually a good thing, in real life. It means that your program runs more efficiently as you increase the number of processors. In other words, it will take less than half the time to run the program if you use twice as many processors. This defies logic upon first examination, but in reality it is possible. Usually it is due to caching and swapping in the CPUs. Also, it usually happens only once, like if you make a graph of number of processors versus the log of the compute time, you see one big dip in time and then things continue on as normal. In my case, the slope of that line is consistently steeper than one should expect.

So I am running a few more runs to see what will happen with more to compute. I'm having it do twice as many computations as before, hoping that maybe this will cause it to average out. Also, I need to investigate because there are two different types of communications on the machine, and it is possible that my low number of processor jobs used the slower communications. I will also see if I can have a meeting with my advisor before I leave for Sweden on Friday.

No comments: