At Fri, 6 Nov 2009 19:31:35 -0500, Thor Lancelot Simon
<tls%panix.com@localhost> wrote:
Subject: Re: git copies of cvs modules available
>
> On Fri, Nov 06, 2009 at 05:51:40PM -0500, Miles Nordin wrote:
> > >>>>> "tls" == Thor Lancelot Simon <tls%panix.com@localhost> writes:
> >
> > tls> silly excuses people are making for the speed and resource
> > tls> consumption issues that users are reporting.
> >
> > ``please make your estimates using git://... instead of http://...''
> > is a reasonable and practical suggestion, not a silly excuse.
>
> If the network protocol used to access the version control software
> can cause an order-of-magnitude difference in the memory resources used
> to check out from the repository, there's something deeply wrong with
> either the design or implementation of the version control software in
> question.
While Git access via HTTP may not be as effective and efficient as it
could possibly be, there are strong reasons why it has not been given
the attention necessary to achieve the ultimate possible efficiency.
So, yes, the _implementation_ may not be well tuned, but nobody serious
about using Git gives a damn, at least not enough to improve it.
The _design_ of the software really doesn't come into play here.
Besides, HTTP is a really horrible protocol for doing many types of
communications, especially at a sufficiently low level. It is not at
all surprising that it cannot be used efficiently in the manner Git
needs to communicate.
I suspect CVS over HTTP could be rather slower and hackish too, though
the CVS protocol is rather simple and small, so perhaps it would do
better. Note though that nobody serious about using CVS has bothered to
implement an HTTP translator for it's protocol either -- at least not
that I'm aware of.
--
Greg A. Woods
Planix, Inc.
<woods%planix.com@localhost> +1 416 218 0099 http://www.planix.com/
Attachment:
pgpKQY8hSsyCj.pgp
Description: PGP signature