2012/5/7 Dimitry T <[email protected]>: > P.S Is there any changes in performance if change in kernel conf i386 to > i686? > "Some reasons why you should not build a custom kernel: > > You do not need to, normally. > You will not get a faster system." > > Can this applies to my question?
Why don't you spend 15 minutes and build two kernels and spend an hour making measurements for your environment and report on the results? If you can't build kernels, then the question is moot anyhow, if you can build kernels, then do it. Apply science and get your answer. Oh, and if you figure it became 0.15% faster at doing $Someting, then do the math of "if I gain 0.15%, for how long do I need to run with this new kernel doing that $Something to regain the time spent thinking about -fzomg-opts, thinking about i386/i686 options in the conf, mailing about kernel settings. waiting for the replies, doing jjs suggested builds and tests". Especially important to get that number right before its time to update the kernel since you get to redo the odd unsupported build again. Seeing how you will not get help for unsupported setups, you may also want to add whatever time it takes you to figure out that a random crash one day may in fact have been caused by odd compiler flags which made gcc use less tested codepaths. How many percents worth is such a crash measured in? -- To our sweethearts and wives. May they never meet. -- 19th century toast
