On Saturday 18 October 2003 11:21, Collins Richey wrote: > On Sat, 18 Oct 2003 10:00:30 +0900 Jason Stubbs > > Try the development-sources first. mm-sources has patches for better > > desktop performance but often breaks things. If you try the > > development-sources first you will at least know what breaks and what > > doesn't before getting even more experimental. > > Just curious. I haven't kept up with the progression in the mm-sources. > Wouldn't a later -testn kernel include most of the stuff in an earlier > mm-kernel? If not, what specifically makes mm- so spiffy? > > I haven't noticed any lack of desktop performance on -testn kernels (at > least 3 months, now running-test8). X responsiveness is normal when > running an emerge or a kernel compile, for example.
I haven't kept up with all the patches, either. -mm however is Andrew Morton's playground for test patches against the kernel. The one that most dramatically improves desktop performance is Completely Fair Queueing (CFQ). Using that I/O scheduler disk throughput will be slightly less but will ensure that no processes get starved for I/O. Try watching a video while doing an emerge with MAKEOPTS="-j 4" and a kernel compile with "make -j 10" or something like that. Without CFQ the video will more than likely become unwatchable. With CFQ it will be barable. I'm not sure what Lovechild did, but with the kernels he put together the video would run perfect. I haven't run -test6 or -test8 but CFQ has actually been pulled out of -test7. Maybe that's why there's no -mm7 at this stage - a little bit of bad blood perhaps? Check out the forums regarding the -test kernels and look for Lovechild's posts in particular. He's summed up the differences between the -test and -mm kernels on several occasions. Jason -- [EMAIL PROTECTED] mailing list
