On Mon, 2004-07-05 at 23:03, Kenneth Porter wrote: > --On Tuesday, July 06, 2004 11:05 AM +0530 peeyush <peeyush@xxxxxxxxxxxxx> > wrote: > > > But now i am wondering if it is worth doing that. Will "rpmbuild SRPM" > > lead to performance advantage over i386 RPMS that are available on the > > site? > > Perhaps, but I doubt it would be significant. Depends on what you're trying > to tune. "Performance" can mean many things. I wouldn't rebuild everything > just in hope that there would be a noticeable speedup. Determine what's > actually slow and optimize just that, if it's possible. Blind optimization > is a bad investment. I agree with this. Where I find srpm most useful is if i want to apply a custom patch, remove a patch Fedora applied, use different build options, or build a newer version than what fedora shipped with. In those cases it usually is best to start with the fedora srpm, read the changelog to see why what was done, and modify from there. As far as optimization is concerned - I prefer to optimize hardware rather than use cpu specific flags at build time. Why? Linux is the only operating system I run that easily will boot a completely different motherboard chipset if my board fries and I need to get back up quickly. This has happened to me, btw - MB fried and i had to be back up ASAP. It was easier than I thought - kudzu had a blast at bootup disabling and re-enabling all kinds of things - but it worked extremely well. Optimizations can screw that. Until there's a way to bundle generic x86 executables with the optimized - I'll just run the generic. 'course generic doesn't mean no optimization, just that it'll run on any pentium class or better (386 and 486 is not an issue for a memory/cpu intensive distro like fedora) I do optimize my kernel - but I keep a generic kernel in grub as well.
