Re: [PATCH] Radix-tree pagecache for 2.5

[Zdenek Kabelac <kabi@i.am>]

Jeff Garzik wrote:
> 
> On Fri, Feb 01, 2002 at 09:06:37AM -0800, Linus Torvalds wrote:
> > Even databases often use multiple files, and quite frankly, a database
> > that doesn't use mmap and doesn't try very hard to not cause extra system
> > calls is going to be bad performance-wise _regardless_ of any page cache
> > locking.
> 
> I've always thought that read(2) and write(2) would in the end wind up
> faster than mmap(2)...  Tests in my rewritten cp/rm/mv type utilities
> seem to bear this out.
> 
> Is mmap(2) only preferred for large files/databases?
 
I've tried to make faster md5summing program and programmed several
ways of accessing file - for the very large files the fastest
way seemed to be  O_DIRECT with threaded precaching.

For fast mmap access I'd to implement two parallel mmpad areas with
madvise MADV_WILLNEED  - then it was almost as fast as read


-- 
  .''`.  Which fundamental human right do you want to give up today?
 : :' :      Debian GNU/Linux maintainer - www.debian.{org,cz}
 `. `'  Zdenek Kabelac  kabi@{debian.org, users.sf.net, fi.muni.cz}
   `-              When in doubt, just blame the Euro. :)