On Tue, 24 Aug 1999, Moray McConnachie wrote:
> > >I'd suggest using a *BSD, ie FreeBSD, as it's filsystem performance is
> > >much better than something like ext2.
>
> People that ought to know tell me:
>
> "No. I don't think even the *BSD advocates claim that *BSD
> beats Linux on large boxes (i.e. SMP and plenty of memory).
> The only places I can think of off-hand where *BSD advocates
> can argue even the possibility for better performance
> against Linux are
> (1) when heavy swapping is going on (which means the box was
> badly balanced when bought especially if it's a compute
> server) and
If your system has resorted to swapping, it's obviously in big trouble.
Linux's comparison problems start long before a system starts swapping.
Given the swap fragmentation problems, there is no "possibility" for
better performance... Even under paging conditions, linux falls far
short. You can't do a linear search through the entire map of the swap
every page you write and expect to maintain any sort of comparison against
some of the more mature unixes. Quite simply, it's flawed and needs to be
fixed; someone will do it sooner or later, but I'm building solutions now.
I'm not suggesting FreeBSD because I'm blindly advocating it's use... ie
"Linux RULEZ"
Which other unixes have you tried?
"Um... Windows???"
Rather, I'm suggesting it because we've tried it and found it performed
better for us than linux. This was not a blind decision; rather an
informed one following many tests and benchmarks.
> (2) certain unusual network uses (due to design differences and
> historical curiosities). There are some functionality differences
> between *BSD and Linux but that's a different issue."
This is not really an issue with filesystem performance. With a table of
2 million tuples, that is what eats the majority of your query processing
time.
I think it's well known that ffs is much faster than ext2. Given
softupdates, it's even better. Softupdates also avoids the filesystem
corruption problems with ext2 because it writes dependant entities in the
proper order, not just any old order they appear in the filesystem cache.
If ext2 did perform just as well as ffs, then why is it being re-designed
for the 3rd time? Maybe ext3 will be better, but for now, I stick with
proven solutions. If it's not broken, don't fix it.
-Michael