OK, Grid2k7 paper was submitted. I was actually surprised with the results. If it does get accepted, I'll probably have to rerun the tests and get some cleaner data points. 5 samples per condition is way too small, but after finding that the tests that had been running for 3 days were screwed up this morning, I had to reduce the sample size.
Still, a 1.5* speedup for doing something as dumb as running a second runahead thread is pretty impressive.
Oh well, now I have to look at the SC|07 paper, which is due in about 48 hours. And for _that_ one, I don't even have the code done, let along the benchmark runs started.