>Is the swapping really hurting performance? On my 512mb system there's >still a bit of swapping when the kernel decides memory can better be >used for disk cache than just for holding unused program data... IMMO yes, if you don't have a good transfer rate from disk :( >Does setting the pio mode _and_ the dma mode make sense? The hdparm man >page says you shouldn't have to fiddle with the -X setting. Is your >transfer rate better after running this hdparm command than before? I think that 2.4.x kernels make some ide optimizations, but yes, before apply the changes the -tT tests give me 84 MB/sec and after 113 Mb/sec. >Have you measured performance increase with those settings? Redhat 8 is >already compiled with i686 optimizations (although code should still run >on older pentium-class machines). This will not give your system the >"boost" some people think it will... I don't understand very well this, only if the optimizations don't affect to binary architecture code and then, the opts. are i585, not for only i686 ( and for an standard rpm i consider i585 ok,but if this software will be used exclusively in i686... ) I can understand that the standard rpm will run 'well' in modern systems, but for example, I'm making tests in my home computer that has Voodoo5AGP and Athlon 1400. If I recompile Xfree without specific optimization options, only recompile, the process detects that I have 3dfx and 3DNow, including both in the recompilation, as newbie I'm, I don't know How much can afect this to my system ( a system with 128Mg and with Video target not well supported due closed hardware specs from a dead company :/ ) A conclusion that I'm getting for now is that is prefereable to compile with -02 against -03 to avoid big executables and preserve RAM to be swapped to disk. But seems to exist good optimization options that don't activate -O2 like -fthread-jumps -fdelayed-branch -fforce-addr or -fomit-frame-pointer. Another question regarding your info about the virtual address space mapped in by the X server. Following you, it's possible that the amount of memory informed in the XF86Config file will be part of this virtual address? In other words, if I have a video card with 64Mb and inform it in the Device Section, when the X server starts, a 64Mb are informed as part of the X process? If it's true, how affect this to the general info about memory used? Thanks and regards Klaasjan Brand <kjb@dds.nl>@redhat.com con fecha 28/11/2002 03:04:22 p.m. Por favor, responda a psyche-list@redhat.com Enviado por: psyche-list-admin@redhat.com Destinatarios: psyche-list@redhat.com CC: Asunto: Re: Testing Psyche - Newbie performance questions > I'had looking in www for tips about performance, but a lot of them > seems to be outdated. > I will recompile all src.rpms, any tip obout optimizations flags? ( > I'm evaluating different combinations and pretend to use -02 -march=i686 > -fomit-frame-pointer -malign-functions=4 -mpreferred-stack-boundary=2 with > gcc 3) Have you measured performance increase with those settings? Redhat 8 is already compiled with i686 optimizations (although code should still run on older pentium-class machines). This will not give your system the "boost" some people think it will... > Any type or advice? Don't waste your time recompiling stuff. The packages Redhat delivers are _almost_ optimized, so there's not much to gain that way. Instead, try leaner software and/or increasing memory. greets, Klaasjan -- Psyche-list mailing list Psyche-list@redhat.com https://listman.redhat.com/mailman/listinfo/psyche-list -- Psyche-list mailing list Psyche-list@redhat.com https://listman.redhat.com/mailman/listinfo/psyche-list
