Re: [RFC] Added editor driver restrictions

Greg Hudson <ghudson@MIT.EDU> writes:

> On Wed, 2003-10-22 at 15:16, Philip Martin wrote:
>> I've added some more measurements to issue 1429 [...]
>> I conclude that your idea may improve ra_local performance,
>> but for ra_dav the server will remain the limiting factor (unless the
>> client is sufficiently "CPU challenged" to make it substantially
>> slower than the server).
>
> By my reading of your numbers, avoiding translation saves over 17% of
> the wall-clock time of an ra_dav checkout, and over 35% of the
> wall-clock time of an ra_svn checkout.

However the proposal is not going to avoid translation, but merely
optimise it, so it's unlikely to do as well as my figures. Also
client cpu becomes less important as client-server bandwith and
latency come into play (my figures were for localhost so bandwith and
latency were about as good as it gets.)

> That's no silver bullet, but
> it's still something to look at seriously, in my mind. I think we're
> all out of silver bullets when it comes to checkout performance.
> (Unless Tobias's latency claim turns out to be true.)

I tried his patch on the issue 1429 test repository that includes
keyword/eol translation, checkout over ra_svn improved by 20% (making
it faster than ra_local in wall clock time!)

-- 
Philip Martin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org