[distcc] Re: fsh

[Jonas Jensen]

On Mon, 2002-11-04 at 19:47, Martin Pool wrote:
> Last time I looked at fsh, it was written in Python.  Much as I love
> Python, I think starting it for every connection may be infeasibly
> expensive.  I haven't measured it though, so I wouldn't write it off.

Just did a quick benchmark on python (not fsh). The python interpreter
takes on average 69ms to start on my Celeron 600MHz. My XP 2000+ can do
it in just 20ms, which is less than the usual latency for sending a
packet over the Internet.
The cost of python may be too great for running over a LAN, but for the
Internet it's perfect. Actually, without support for this distcc
shouldn't even be run over the Internet, so this is enabling a whole new
way of using distcc.

Also, you wouldn't really be adding fsh support, but just support for
any rsh compatible program, including any future fsh-style program that
may be written in C. And even plain rsh can be (a bit) more secure than
what distcc does now.

Another reason fsh support would be nice is that it would enable
compression, because ssh supports that. Lack of compression is another
reason why distcc can't be used over the Internet today (at least by
normal people like me who don't have 10Mbit pipes).

> Using fsh would be nicer (lazier) than writing custom
> connection-caching code.

If you do decide to write custom connection-caching code some day, you
could just code it to be rsh-compatible, which would make it useful for
everyone who uses fsh today, and it'll make distcc less bloated (as you
advocate yourself in your docs about why you don't merge with ccache).

/Jonas Jensen