[nycbug-talk] benchmark OS perf

[Hans Zaunere]

> > I know this is just chum in the water for you sharks..  but still an
> > interesting article.
> Just another test made by a clueless person.
> 
> One doesn't test performance of an OS by compiling and running a certain
> package. You get "how well does mysql run on this OS" (article body says
> that, article header does not).

Agreed - the article's heading is a bit superfluous.

> Author has no idea what 'cold cache' and 'hot cache' is. Nowhere he says
> whether he rebooted *just before* running the benchmark itself. Since he
> doesn't say that, the results are useless - you can't compare benchmarks
> with hot cache with cold cache.

True, and it's a little bogus that we have to wait for the actual results.  Seems like it'll generate more ad revenue for newsforge, though.

That said, PeterZ is good stuff, and I'll bet he guided this guy well.  This article does raise a major point, though, and that's threading under FreeBSD, and the BSD's in general.  Speaking specifically to FreeBSD, it's hard to get it be a creditable OS for databases when all you can say is "well, pick a thread library; most should work about the same"

After talking with a FreeBSD core developer, they are working on this for 5.3.  And, in fact, the thread project itself uses MySQL to test their thread library (not the the whole OS) for performance, leaks, etc.  So, the future is bright...

H