BFS Scheduler Benchmarks
Written by Michael Larabel in Software on 14 September 2009 at 04:00 AM EDT. Page 3 of 4. 105 Comments.
Of the tests we ran when looking at Con's scheduler, one of the drops against the Completely Fair Scheduler was found with the 7-Zip compression test, where its MIPS count had decreased. However, as you can see from the graph, the loss against CFS was very small.
GraphicsMagick is driven by OpenMP for its multi-processing support and with the image resizing benchmark the performance under BFS was better by about 7%.
The biggest performance boost we encountered when testing out the Brain Fuck Scheduler occurred when carrying out static web-page serving performance via the Apache web server. The number of requests per second that could be sustained was 3140 with BFS and 1902 with CFS, or a 65% boost in performance.