On Dec 6, 2008, at 10:44 AM, Ross Walker wrote: > > On Dec 5, 2008, at 7:18 PM, Rainer Duffner <rainer at ultra-secure.de> > wrote: > >> >> Am 06.12.2008 um 01:02 schrieb Ignacio Vazquez-Abrams: >> >>> On Fri, 2008-12-05 at 23:57 +0000, Michael Holmes wrote: >>>> 2008/12/5 Matt <lm7812 at gmail.com>: >>>>> I have a server running Centos 4.7 32bit. Will moving from 4Gig >>>>> of >>>>> RAM to 8Gig do any good? Since its 32bit I assume it will only be >>>>> able to address the first 4Gig not? >>>> As long as you are using a SMP kernel you can use up to 64GB of RAM >>>> (though each proccess can only address 4GB of this). So if you can >>>> find any trace of "SMP" in the uname (grep is your friend) then it >>>> should work fine. >>> >>> PAE, not SMP. >>> >> >> >> He should be able to replace the kernel via rpm -e and rpm -i >> >> That said, I doubt he'll actually see a benefit. >> PAE is slow. >> If you want to see a real performance-gain, install 5.2 x86-64. > > He'll see a benefit. PAE slowdown is humanly unnoticeable for short- > term transactions, it's difference is in high nano seconds or low > micro seconds. > > All 5.0 kernels are PAE by default. > > So are WinXP_SP2/Win2K3/Vista/Win2k8 and Mac OS X kernels. > > Only go 64-bit if it's completely necessary. I see that I had mis remembered some changes in 5. At least with regard to the upstream provider, on X86 the desktop version has a limit of 4GB of RAM, regardless of how much more memory you have. And they removed the hugemem version, so instead of up to 64GB of RAM on 32 bit, you can only get to 16GB for server versions. I was combining the bunch, and thinking that you could now only get to 4GB on 32 bit. Of course, I believe that XP SP2 is now limited to 4GB of RAM on 32 bit OS's, less when you factor in device space, due to problems with some device drivers.