Hey all,
I'm still working on sort, now as a part of a group project in Professor
Eggert'sclass, and I wanted to show you guys some prelim results.

Tests are conducted on an otherwise free 8 core machine (Xeon X5460,
2 dies of 4 cores each). The input file is 1 million lines, 96 MB, and a
buffer size of 500M is specified.

>From Glen's patch last year, we had elapsed times of
1 thread: 5.59 s
4 threads: 2.34 s
8 threads: 1.94 s

Our results are:
1 thread: 5.15 s
4 threads: 2.05 s
8 threads: 1.65 s

The main speed up in the 1 thread case is because we moved
output to the top level of merging. At any rate, expect a patch
submission in a few weeks.


Reply via email to