Re: [PATCH] Re: 2.1.xxx makes Electric Fence 22x slower

Alan Cox (alan@lxorguk.ukuu.org.uk)
Mon, 7 Sep 1998 14:37:17 +0100 (BST)


> Such multiple algorithm dynamic schemes have already been shunned by
> most of the people on this list early on in the thread. It's what
> Solaris does, and it's not only complex, it's also stupid. Find one

Do I see religion creeping into the discussion at this point. People have
been saying lots of things Linux does are stupid too.

> algorithm that works nicely in all cases, so we don't have to maintain
> a piece of code which uses two different schemes.

There are three cases of interest

o Low number of VMAs (typical application)
o Medium number of VMAs (very rare)
o Large number of VMAs (electric fence, lpmud, texas..)

and the large number of VMA case is the important - you want a quantifyable
worst case for things in an OS kernel.

Right now case 1 is covered, case 2 is acceptable, case 3 is hopeless. The
split list/avl implementation also covers case 3 well for very little new
code - maintenance here is of code thats already well tested.

Fuzzy hash is optimised for the rare case, is worse at both the typical
application and worst cases and because the hash is published has a
terrible worst case for anyone who is bored enough to write programs to
demonstrate it.

I'm sure fuzzy hash is good for some things, this just isnt one of them

Alan

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/faq.html