Re: TCP/IP one->many

Adam D. Bradley (artdodge@cs.bu.edu)
Thu, 8 Oct 1998 14:13:27 -0400 (EDT)


On Thu, 8 Oct 1998, Tom Mills wrote:

> Bob Lanning wrote:
>
> > look into multicast. (not TCP) TCP requires 2 end points for every
> > connection. There are a bunch of negotiations (window,mtu...) that go on for
> > each connection. These negotiations are done throughout the existence of the
> > connection.
>
> I've thought about UDP but don't want to incur the overhead of making sure the
> packets are delivered reliably, in order, etc. Also, with UDP isn't there router
> considerations if running over a WAN ? The application I am writing must handle
> heady I/O load and be reliable. I was thinking that one could add a thin driver
> to intercept TCP writes and handle the many send there given the appropriate
> target IPs by the user application.

Well, you really have 2 options:

1) Have some sort of wrapper library, so your application is presented
with a "write to one socket" interface which gets transformed into writes
to a whole bunch of sockets (thus you incur all the bandwidth and TCP-code
overhead of just sending it to all of the clients individually anyway)

2) Look into one of the "reliable multicast" technologies (SCE, RMP, RMTP,
RMTP2, RAMP, etc). I seem to remember there being lots of stuff on these
somewhere in the tasc.com website. SCE is probably most (conceptually)
like what you want.

Adam

--
You crucify all honesty             \\Adam D. Bradley  artdodge@cs.bu.edu
No signs you see do you believe      \\Boston University Computer Science
And all your words just twist and turn\\    Grad Student and Linux Hacker
Reviving just to crash and burn        \\                             <><
--------->   Why can't you listen as love screams everywhere?   <--------

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/