[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: [PATCH] Saving a few cycles, part 2/2

From: Hyrum K. Wright <hyrum_wright_at_mail.utexas.edu>
Date: Mon, 26 Apr 2010 19:32:19 -0500

On Mon, Apr 26, 2010 at 7:11 PM, Stefan Fuhrmann <
stefanfuhrmann_at_alice-dsl.de> wrote:

> Julian Foad wrote:
>> On Sun, 2010-04-25, Stefan Fuhrmann wrote:
>>> here comes the second set of local optimizations.
>>> It contains a number of simple improvements over
>>> the original code.
>> Hi Stefan. Thanks for working on optimization. Saving a few cycles can
>> be useful if it's in frequently-executed code so that the overall gain
>> is significant. At the same time, we are trying to keep a good balance
>> against code simplicity. Much of our code is more complex and longer
>> than it should be already, so I want to be careful about any change that
>> makes it even longer.
>> Have you got (or please could you get) some measurements for these
>> optimisations that you could show us? From just looking at the patch, I
>> find it hard to believe that some of the changes make a consistent
>> difference, at least from the perspective of thinking about different
>> compilers. Specific comments and questions are below.
> All this code is over 3 weeks old now. I simply sent
> the Easter weekend optimizing SVN, APR and zlib.
> The total performance gains are tremendous.

Stefan, I've been following the discussion but haven't had too much to
contribute. I know there's been some good feedback which has led to some
interesting technical discussion. I'm personally thankful that somebody is
doing this kind of analysis and looking at these types of places that
Subversion can be improved. Thanks for being patient with the community as
we digest your suggestions.

> Now, I'm in the tedious process of slicing the changes
> into smaller, easily digestible chunks. Once I got them
> all posted, I plan to write an extensive analysis of the
> current SVN performance including several measurements
> for different OS, FS, patch levels, repository formats etc.

A report I'm anxious to see!

> Bottom line: Subversion-sized projects could technically
> be checked out in less than 1 second (wire-speed permitting).

That would be awesome, especially since many of our users are accessing
repositories on local networks. There are many pieces to the puzzle of
course, but we should be saturating the network, because it's generally the
scarcest resource.

Received on 2010-04-27 02:32:45 CEST

This is an archived mail posted to the Subversion Dev mailing list.