> From: Karl Fogel [mailto:kfogel@newton.ch.collab.net]
> Sent: 22 May 2002 23:50
> Greg Stein <gstein@lyra.org> writes:
>> That algorithm "won't work" ... there is too much searching taking place.
>> The current pools code is way fast because it doesn't have to search for
>> blocks in the typical allocation case.
>>
>> The pools code is quite sensitive. It is noticable if you add even one more
>> 'if' statement to the typical-use codepath.
>>
>> (the proposed patches don't seem too bad because they really only come into
>> play at non-typical points: when you need a new block, and when you're
>> freeing a pool)
>
> Ah, okay, so the "active block" means "try me first no matter what",
> and the "inactive blocks" are "try us before allocating a new block",
> and the distinction is made for speed.
*nod*
> (Except that even with the patch, we'll only try the first of the
> inactive blocks.)
Yes, the first is the one with the largest free space left, since we
keep the list ordered in a simple manner. There is room for improvement
in that department. A faster way of keeping the list ordered can be
coded up. A priority queue might even be a better idea, since we only
need to have the head of the list be the block with the largest amount
of free space.
> Hmm.
>
> I guess I don't know how the benchmarks look, so I'll shut up now :-).
*grin*
Sander
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu May 23 00:37:29 2002