Slow "bzr branch" on Savannah

[Eli Zaretskii]

> Date: Fri, 18 Feb 2011 18:04:17 -0600
> From: John Arbash Meinel <john at arbash-meinel.com>
> CC: bazaar at lists.canonical.com
> 
> I read it, and was thinking about it. I still have it marked "unread" in
> my folder (the last one, in fact). I just haven't had time to really dig
> into it and determine the implications.

That's okay, it can wait.

Thanks for considering how to improve this.  I really think this is
important for having bzr scalable for large projects with a long
history.

> The immediate take away was just that using the smarts on the client
> side seems to work a lot better than taxing the Savannah server for
> computing what to send. And I can sort of see that being true for "fetch
> everything". If the server is loaded, then offloading all the
> computation to the client works well for CPU. And if you are fetching
> everything anyway, there isn't a loss on the client side for fetching
> more than it needs (because it needs everything).

I would be interested in seeing the results of using nosmart+ URL by
Ulrich (CC'ed).  Since I observed it to be a win only on a machine
that sits on the same LAN as bzr.savannah.gnu.org, I hesitate whether
to update the wiki at http://www.emacswiki.org/emacs/BzrForEmacsDevs
to advise people use nosmart+ on the first "bzr branch" command.  But
if there's a similar speedup on remote machines, I will make that
change on the wiki.

> I would imagine that all future updates would be worse without smart
> server support.

Yes, disabling the smart server after the initial branching is
certainly out of the question.  It works very well for routine day to
day ops.

Thanks.