HG clone network performance
Mathieu Lacage
mathieu.lacage at sophia.inria.fr
Sun Dec 30 11:54:25 UTC 2007
On Mon, 2007-12-17 at 13:40 +0100, Roman Kennke wrote:
> Hi,
>
> > On Dec 17, 2007 5:17 PM, Roman Kennke <roman.kennke at aicas.com> wrote:
> > > Yeah sure, I considered that. I'm not talking about the actual amount of
> > > data transferred (this is ~300MB for the HG vs. 200MB CVS data). I was
> > > only wondering, why HG doesn't saturate the line when cloning. I wonder
> > > if there's room for optimization here?
> >
> > I am quite positive a multi threaded approach to pump more data over
> > the network for bulk data transfer might help. I am quite new to
> > mercurial core to be able to suggest anything at a low level.
>
> I'm not the network expert, but why/how should multithreaded
> datatransfer help with network performance? I mean, if one thread could
Well, you could probably get the same result with a single thread and
select. The crux of the proposal here is to create multiple tcp
connections to the server. Given tcp's congestion control algorithm,
this can be a win but, as pointed out by matt, it is a bit like
'cheating'.
Mathieu
More information about the Mercurial
mailing list