: I think I can probably find some resources to put a linux-cvs.sgi.com
: machine on the net. We have some spare ports on our internet ethernet.
: If I put a machine there with disk space, and provide logins, can you
: guys do the rest?
With sufficient privs, yes. It will be done.
: Note that I might be putting a linux pc there and
: hacking it to limit the rate at which the data can move. Has anyone
: done this? That's SGI's primary concern, they would like to "reserve"
: some bandwidth for other stuff on that T1 line.
Hmmm, ordinarily you'd just rate limit FTP. However, since this will
be used for CVS operations, etc, etc, that will likely not be enough.
It would have to be done in the driver for the ethernet card. I've
not done this before on a linux machine, but I think it would be
relatively easy to pretend the packets take a really long time to
transmit and somehow tie into the clock interrupt (likely via a
timeout that gets armed in the transmit complete interrupt that most
ethernet cards give). I'd imagine that would be the easiest thing to
do. I'd not worry about what packets want to go out, just do it for
all of them. That should give the least pathological behavior.
I don't think this project would generate that much of a load on the
T1 to warrant effort to implement the throttling. Therea aren't that
many of us, last time I checked. Hey FTP.FNET.FR guys! How much
bandwidth (file transfers, etc) is used now? If only the developers
of Linux/MIPS had access to this machine, then it would be much less.
The linux kernel at T1 speeds represents about 30 seconds of T1 time.
Generally development activity will push accross patches that are much
smaller than that. There are about 10-15 linux mips developers right
now that are active.
However, it is your machine and your bandwidth, so do what you think
is necessary for this to be allowed to happen...
Actually another idea springs to mind. It may be a simple matter to
hack the default TOS to be really horrible. That is if your gateway
routers do the right thing with TOS (given that you are SGI, and who I
know has worked on the TCP/IP for your SGI boxes, the chances are good
that this will do the right thing). That way the packets from the
linux box will always be last in the queue.