Re: 1.7 timing tests: update great, checkout needs work, upgrade horrible

On 12/09/2010 12:03 PM, Hyrum K. Wright wrote:
> This feels really bizarre. Isn't checkout just "create an incomplete
> directory, then update it"? As such, one would think the performance
> would be comparable. Bizarre indeed....

Huh?

Blair doesn't state whether or not he's talking about "no-op updates" --
that is, an update on an already up-to-date working copy -- but that's what
I'm assuming.

The primary benefit of 1.7 for updates is that we don't have to make a full
crawl just to lock the tree. So it's nice to see that Blair's client is
independently confirming that yes, things are better here.

But while checkout is implemented as an update under the hood, that sort of
an update is far, far cry from a "no-op update". The benefit we gained in
1.7 for updates I mentioned above -- not having to crawl the tree to lock it
-- doesn't even apply in this scenario because there's no tree to lock (it's
a single empty directory, right).

So, yes, it's perfectly reasonable that we should see the performance as
compared against 1.6 of these two effectively very different operations
trend in different ways -- even opposite directions.

-- 
C. Michael Pilato <cmpilato_at_collab.net>
CollabNet   <>   www.collab.net   <>   Distributed Development On Demand