Messages in this thread | | | From | Steven Bonneville <> | Subject | Re: HINT benchmark and Linux | Date | Tue, 23 Mar 1999 09:26:02 -0600 (CST) |
| |
> To grossly simplify, this benchmark purports to measure performance > a wide range of problem sizes. And sure enough, when run on FreeBSD > the performance curve shows the classic ``3-step'' drop-off when the > data size moves from cache to main memory to paging. The Linux > graph is similar but with a much less clearly defined ``shape'' > and the small-problem peak performance is about 15% less that FreeBSD. > (See http://www.cs.washington.edu/lab/sw/HINT/hintgraphs.html.)
I've tinkered with HINT on some boxes here as well; we haven't tested it on a box running 2.2.x, but we have on boxes running 2.0.35 and 36.
Your graphs for Linux look low compared to our results. What net MQUIPS did you get for double? We observed net MQUIPS of 18.1 for double on a single PII-400 running 2.0.35 (128 MB RAM).
We've got a good "three-step" curve of those results graphed online:
http://www.ima.umn.edu/stats/hint/intel/violet/
although the double results look a little crushed due to scaling effects resulting from being on the same graph with the integer results.
Steve Bonneville IMA System Administrator
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |