hi..
> What! That makes no sense. A compiled kernel is different from a
> non-compiled one? Someone compiled it once. He said he didn't make any
> changes to the FreeBSD kernel except recompile. I will be shocked to
> hear he sees any difference in a kernel recompile.
no, what i'm saying is that the options (which are legion) chosen when
compiling a linux kernel affect its performance. this is obvious when you think
about it for a second, bruce.
will a kernel with 18 uneeded options (such as VFAT FS support, etc,etc) that
adds an extra several dozen K to the kernel and which was compile for a 386 be
slower or faster than one that is compiled with only the options you actually
USE and is compiled to take advantage of the instruction sets on your juicy
PentiumIII? the former is what you get with the stock redhat kernel, the latter
is what you get when you compile it yourself...
to take it to another arena: which would be faster: postgres installed from a
binary package or postgres compiled from scratch with exactly what you need
optimized for you chip?
>
> That said. we are tracking down an linux OS kernel problem today in
> pre-7.0 where read-aheads are turned off when a seek is performed on the
> file. I have downloaded the linux kernel 2.2.0, and I see the cause is
> that only repeated reads/writes without a seek cause file system
> readahead.
2.2.0 isn't the stable kernel =) 2.2.13 is... other decent choices are 2.2.5
and 2.2.10 (though if using 10,. you may as well use 2.2.13).. the only reason
for using 2.2.5 (as we do) is for driver suport.. in our case, the driver
available for one of our onboard disk cards is written for 2.2.5 and 2.0.36
only (stupid dpt)
>
> Other OS's check to see if a previous read-ahead was used, and control
> read-ahead that way. Much better. Or am I making an eroneous
> comparison? This was a real-world test where someone is saying pre-7.0
> is slower for sorting.
yeah, i've been following that one too.. what filesystem are they using
(ext2fs?) and what version of the kernel? if using 2.2.0, then that would
might be the reason...
>
> Are you telling him not to say anything about what he sees?
no, not at all... what i _am_ saying is that if you are going to say something,
or even make observations for yourself, you and everyone else is best served
when the observations hold water... its the scientific model: a poorly designed
experiment results in poor results. his experiment (not him, of course.. i'm
sure he's a great guy) was poorly constructed. this leads to innacurate and
deceiving results..
and i'm not "sticking up" for linux.. merely accurate experimentation.
think of mysql's crash_me test. how good of a test is it really? and therefore
how good are the results?
--
Aaron J. Seigo
Sys Admin
************
