[haiku-development] Re: More fun with numbers [was: Networking speed update]

Stephan Assmus <superstippi@xxxxxx> wrote:
> It will be most interesting to repeat these tests when the new I/O 
> scheduler is effective! :-D

I doubt this will gain a lot of speed, actually, at least not at first. 
The I/O scheduler alone will only bring improvements when more than one 
thread is using the hard drive (which is not the case for SVN).
It will just create a framework that allows us to do other 
optimizations that might bring more speed :-)

However, I guess most of the work is done by the block cache when SVN 
is at work -- and that one doesn't do any read-ahead either with the I/
O scheduler.

"Rene Gollent" <anevilyak@xxxxxxxxx> wrote:
> That would be because svn status is most heavily hitting directory
> blocks, which BeOS and/or Zeta don't cache at all iirc.

BeOS/ZETA cache file, directory, and meta data in the same way using 
the block cache. The only problem of that cache is that a) it has a 
very limited block count, because b) it's not integrated with the VM at 
all, and c) it has a single lock for all volumes using it, and d) since 
it is block based, it doesn't know what blocks to load (it does 
actually read more blocks than you ask for (synchronously), but this 
could be any data).

Bye,
   Axel.

Other related posts: