Some timings using bzr
david.ingamells at mapscape.eu
Fri Jul 11 09:13:59 BST 2008
Andrew Cowie wrote:
> ???I'll leave the actual thrust of your investigations, NFS performance,
> to others, but to
> On Fri, 2008-07-11 at 08:21 +0200, David Ingamells wrote:
>> 1. file:// is better than bzr:// when using a local destination.
>> Surprisingly bzr:// across the network was quicker than bzr://
>> used on the same machine.
> I can't really say as how this is surprising; bzr:// is a network
> transport and is designed for the resource profile (latency, etc)
> inherent in doing operations over physical networks. Not much point in
> putting a TCP/IP stack & a network protocol in the way of doing local
> read and write.
> If you characterize your second sentence as: "unsurprisingly, file://
> URLs are more appropriate for local disk operations than bzr://" it
> makes perfect sense.
> One of the things that makes Bazaar (and 3rd generation DVCS in general)
> interesting is that it is incredibly difficult to do apples-to-apples
> comparisons _because_ so many things are different depending on what
> you're doing. This really shows up in http:// vs bzr:// comparisons; for
> some things one is (unintuitively) faster than the other (say, pulling
> lots of data), but the next minute you do something else (say, pushing
> varied revisions) and the performance characteristics are *totally*
> different. So I respect the people working on this project; they appear
> to have a fair bit of discipline about measuring things that are
> actually relevant.
Read the snippet again! Maybe you'll have some caffeine in your blood
supply this time.
I DO find it surprising that bzr:// to the same machine is slower than
the same command to a different machine, especially one with a much
slower network link. I TOTALLY agree that file:// will be much quicker
that bzr:// when no network is needed - which is what I said! But that
is NOT what I was talking about in the sentence that starts "Surprisingly".
I also respect the work of the bazaar team: this was in part why I chose
bzr as our tool when I evaluated the alternatives in January this year.
I did not intend my original message to imply any criticism. I did the
tests to try to understand why my team is having horrendous performance
problems which I have proven with my tests are to a large extent caused
by non-bzr issues. My intention was to share this experience with the
bzr community at large and possibly help the bzr team a tiny little bit
in their efforts to make bzr even better.
From what I have seem of the recent discussions on the forum I expect
that version 1.6 will show better results.
So great work, guys and a great tool!
More information about the bazaar