🚀 go-pugleaf

> > I have about 5 AT&T 3B2/300s linked together over 3BNET (ack!).  
> > I want to know if their is any way that I can dump my 3B2 to a
> > tape drive hanging off of our vax.
> Hmm, at 9600 baud, it would an awful long time to backup 30 Meg :-).
> . . . 
> What would really be nice is to have nisend be able to interactively
> read stdin and write across the net so you could do something like this:
> 
> 	find . -print | cpio -ocm | nisend -d vax -f "vaxtape"
> 
> If you could do this, no disk space problems would present themselves.

I have written a pair of programs "sender" and "receiver" to implement 
exactly this.  The stdin of sender becomes the stdout of receiver; the
two programs run on different machines and communicate via "raw" ethernet.
Thus, I can back up machine borel using the tape drive on abel via

borel:  find whatever -print | cpio -oac | sender
abel:   receiver >/dev/rTheTape

On a lightly loaded net I can transfer about 2 meg/minute, which is faster
than our puny cartridge tape drive can write even at the best of times.
At 9600 baud, the transfer rate would be ~60K/min.  The bottleneck for me, 
as it turns out, is the raw device interface to the tape drive, which can only
crank out 1 meg/9 minutes, or ~3 hours per 20+ meg tape cartridge.
The AT&T tape backup utilities bypass this raw device driver and turn on
"streaming" mode to write at 1 meg/1 minute; source for these utilities,
of course, was *not* included in our source distribution.

Release 3.0 of UNIX is supposed to allow one to use devices on remote machines,
so I could do the following on borel (referring to the AT&T utility ctccpio):
    ctccpio -oa -T abel!/dev/rSA/ctape1
(I think I have the syntax right.)  We have ordered 3.0, but it has not yet
arrived.

My implementation of sender and receiver is a little sloppy, especially
since I started with some major misconceptions of just how ethernet worked.
It also prints a bunch of dumb diagnostic and timeout messages as it 
proceeds.  Thus, I'm reluctant to post it until I can clean it up.
However, I will send it to anyone who wants it; it seems to work and, since
it deals in raw ethernet packets, it should be useable between any pair of
ethernet systems (not just 3BNET).  The basic idea is that sender sends a 
1K data packet, and receiver sends an acknowledgement, and so on until done.
At that point, things hang.

                                            peter dordal
                                            loyola univ. of chicago
                                            dept of mathematics
                                            6525 N. Sheridan Road
                                            Chicago IL 60626-5387

Send requests to the address above or to
                                        ...!ihnp4!gargoyle!cantor!pld