On Tue, Apr 04, 2006 at 03:25:24PM -0700, Miles Keaton wrote:
> This is a serious question, for heavy users of OpenBSD in
> big/production/heavy-traffic situations.
> 
> For years, our small company used OpenBSD for *EVERYTHING* because I
> personally prefer it.   (We run a pretty popular database-driven
> website.)
> 
> All mail servers, web servers, database servers, were all OpenBSD.
> 
> But then some threads-issue with MySQL on OpenBSD made us switch to
> FreeBSD for our database server, in an emergency.  The increasing load
> on the server was making OpenBSD buckle, and switching to FreeBSD (on
> the same hardware!) was a 100x speed improvement.  Unfortunately, we
> switched other servers to FreeBSD, too, to standardize, and have been
> almost entirely FreeBSD, since.
> 
> Ah, but this was back in 2001 or so.  I know things in OpenBSD are
> better now.  SMP.  Etc.
> 
> Things at our company have grown enough so that we finally have
> load-balanced servers, so not all traffic needs to be whomping a
> single server.
> 
> We're setting up some new hardware, and I want us to take a look at
> OpenBSD again for things like webservers and database servers.  (Not
> too happy with the SMP in FreeBSD.)  Maybe even get back to our old
> situation of being 100% OpenBSD for everything.
> 
> Which leads me to my real question for you heavy users of OpenBSD in
> big/production/heavy-traffic situations:
> 
> When would you NOT use OpenBSD?
> 
> When would you choose one of the other *nix over OpenBSD?
> 
> Is OpenBSD appropriate for a busy webserver or super-loaded database server?
> 
> I've seen old "O.S. shootouts" benchmarks comparing O.S.'s and often
> showing Linux or FreeBSD excelling at webserving or
> database-performance, but I don't know if that's just old data or the
> benchmarkers didn't have OpenBSD tweaked right.
> 
> As you can tell I'd *like* to go back to OpenBSD-everywhere but
> thought it would be wise to ask the misc@ gang about this first.

OpenBSD just isn't the fastest around (partly due to the abundance of
crypto, and partly due to the fact that it just isn't that much of a
priority). It also does not have the best multiprocessing or threading
support around.

Additionally, hardware support is not as good as from, say, Linux. Or
even FreeBSD. And if the hardware is not supported, an OpenBSD box
doesn't do a whole lot of good.

Finally, some programs are written in a sufficiently Linux-centric
fashion that porting them is rather difficult. Some have been
succesfully ported, but there are not quite as many packages for OpenBSD
as for, say, Linux. Or even FreeBSD. (Seeing a pattern here?)

SMP is supported, but some other OSes do a better job. If you are
sitting on a 16-way UltraSPARC box with many gigabytes of memory, you'd
probably be better off putting Solaris on it - that's what it's for. (Do
put an OpenBSD firewall in front, though.)

Also, OpenBSD excels at security and code correctness. If you do not
connect to the internet and do not develop OS-level software, neither
count for much. A gaming box in the basement, without internet
connection, would not see that much benefit from OpenBSD - dump Windows
on it. And live with the occasional BSOD.

Pretty much the same is true for some webserver/database server
situations where the last few percents of performance are really
necessary.

However, I do think most of the really inefficient stuff has been
improved sufficiently that the differences are not likely to be that
big. But the userspace thread implementation still has its limitations.

                Joachim

Reply via email to