MARKUS@LOYVAX.BITNET (01/20/88)
Hi,
We have phone here, but what is more commonly used is SEND and a
program called SAY. I'm assuming SEND is implemented on most nodes and
could be used much as here (ie. for sending one liners but not through mail).
I'd be curious to see a comparison of system resource usage between phone
and SEND (or any other common one line message facility).
mark plummer
MARKUS@LOYVAX.BITNET
loyola college (baltimore) acsCLIP08@VAXD.STRATH.AC.UK (02/10/88)
Has everyone noticed how inefficient at communicating PHONE is? if you open a channel to a PHONE mailbox while it is in use, and read from the channel...you will see that every character typed by a use is sent preceded by their nodename and username in other words...the ratio of bumpf to actaul message data is less than 10 to 1
jeh@crash.cts.com (Jamie Hanrahan) (02/18/88)
In article <8802150517.AA22133@ucbvax.Berkeley.EDU> CLIP08@VAXD.STRATH.AC.UK writes: > >Has everyone noticed how inefficient at communicating PHONE is? >if you open a channel to a PHONE mailbox while it is in use, >and read from the channel...you will see that every character >typed by a use is sent preceded by their nodename and username > >in other words...the ratio of bumpf to actaul message data is less than >10 to 1 (What is ``bumpf''? From context I interpret it as ``total bytes sent'', but why ``bumpf''?) I wouldn't worry overly much about this. The time required to send a message through a mailbox (and to read it) is pretty much independent of the message size for messages under 256 or 512 bytes in length. (It could be worse. Why, I've heard of some systems where EVERY CHARACTER sent through an interprocess communication channel required a complete trip through the I/O subsystem!) Things get better if your system is busy. When PHONE's terminal read AST executes (because you typed a character) it does a timed read with timeout period 0 and a large buffer; this gets as much as possible (up to buffersize) out of the terminal typeahead buffer. So if your system is busy enough that you can type several characters before PHONE reads the first one, PHONE will slurp them all in, and then send them with a single mailbox write... much the way the input silo is used on terminal muxes. The nifty thing about this trick is that the busier your machine is, the better it works (because more characters have time to build up in the typeahead buffer). A while back (when all the ``PHONE is a horrible waste of system resources'' messages were floating through this newsgroup) I ran a little test. I used autorepeat to send characters through PHONE (at about 25/second) to another terminal, while doing a MONITOR MODES on a third terminal. With nobody else running I had about 70% of my 8200 still idle. 30% of a 780-class machine may seem like a lot, until you stop to consider just how many average typists it's going to take to generate an average of 25 keystrokes per second in real life...