JBS%DEEP-THOUGHT@mit-eddie.arpa (03/06/86)
I have been experimenting with running a vt220 here at 19.2Kbaud. The terminal works great except when sending more that about two pages without a pause--at that rate, the terminal can't quite keep up and loses a little data. The lost data is not really my problem though. What I'm worried about is that on the console, the system spits out occasional "dz0,4: silo overflow." Does anyone know what is wrong, and if so, what I can do to fix it? (oh, did I forget to mention that it is running 4.3BSD?) Jeff Siegal -------
gwyn@brl-smoke.UUCP (03/07/86)
In article <1577@brl-smoke.ARPA> Jeff Siegal <JBS%DEEP-THOUGHT@mit-eddie.arpa> writes: >The terminal works great except when sending more that about >two pages without a pause--at that rate, the terminal can't quite >keep up and loses a little data. > >The lost data is not really my problem though. What I'm worried about >is that on the console, the system spits out occasional "dz0,4: silo >overflow." Does anyone know what is wrong, and if so, what I can do >to fix it? The first paragraph makes it apparent that you have not enabled DC3/DC1 flow control. As you overrun the terminal's buffer, it will keep sending DC3s to the host to try to get it to stall. These are probably filling up the DZ11 silo faster than the host is able to empty it. If the DZ11 device driver relies on input silo alarm rather than per- character interrupt with short-circuit lookahead before RTI, then there is not much safety margin. If you enable flow control, the problem should go away.
jack@boring.uucp (Jack Jansen) (03/08/86)
[Sorry for posting, but I couldn't reach the author by mail...] In article <1577@brl-smoke.ARPA> JBS%DEEP-THOUGHT@mit-eddie.arpa writes: > >The lost data is not really my problem though. What I'm worried about >is that on the console, the system spits out occasional "dz0,4: silo >overflow." Does anyone know what is wrong, and if so, what I can do >to fix it? Well, in fact you should even be grateful if it runs at 9600 baud... The DZ is completely interrupt driven, which means that *every character* causes an interrupt. So, 8 lines at 19.2Kb can produce an awful amount of interrupts...... Whenever the VAX doesn't serve input interrupts in time, the input silo fills up with unprocessed characters, and eventually some characters will be dropped. There are two solutions: - Set speed to 4800 baud. You will also see the performance of your machine jumping up, since it doesn't have to spend so much time on interrupt handling anymore. - Replace the DZ by something better, probably some DH-11 clone. These things use DMA for output, so you get far less interrupts. -- Jack Jansen, jack@mcvax.UUCP The shell is my oyster.
jbs@mit-eddie.MIT.EDU (Jeff Siegal) (03/09/86)
In article <1614@brl-smoke.ARPA> gwyn@brl.ARPA writes: >The first paragraph makes it apparent that you have not enabled DC3/DC1 >flow control. As you overrun the terminal's buffer, it will keep >sending DC3s to the host to try to get it to stall. These are probably >filling up the DZ11 silo faster than the host is able to empty it. If >the DZ11 device driver relies on input silo alarm rather than per- >character interrupt with short-circuit lookahead before RTI, then there >is not much safety margin. > >If you enable flow control, the problem should go away. Nope. I intentionally did not enable flow control on the VAX, because in order to have functional ^S and ^Q keys, flow control must be off on the terminal (yes, I know this is a misfeature). I can't disable the ^S and ^Q keys (as turning on flow control would do) since we need them for Emacs. In other words, the terminal is NOT sending DC3's at all (unless I press ^S). I can't figure out why the dz-11 thinks data is being sent to it; sounds like a Unix bug to me. BTW: Does anyone know if I can set up hardware flow control (DSR/DTR) with this equipment? Jeff Siegal - MIT EECS
jbs@mit-eddie.MIT.EDU (Jeff Siegal) (03/09/86)
In article <6815@boring.UUCP> jack@mcvax.UUCP (Jack Jansen) writes: >Well, in fact you should even be grateful if it runs at 9600 baud... >The DZ is completely interrupt driven, which means that *every character* >causes an interrupt. So, 8 lines at 19.2Kb can produce an awful >amount of interrupts...... I am only attempting to run one line at 19.2Kb (mine). Most of the other lines are either connected to 1200b modems, or not used. All of the users access the machine via a network. One terminal, at whatever speed, doesn't seem like that many interrupts. Also, see below about interrupts. >Whenever the VAX doesn't serve input interrupts in time, the input >silo fills up with unprocessed characters, and eventually some >characters will be dropped. There are two solutions: >- Set speed to 4800 baud. You will also see the performance of > your machine jumping up, since it doesn't have to spend so much > time on interrupt handling anymore. Again, since there is only one fast terminal on the VAX, this shouldn't be necessary. I have looked at the number of interrutps on the system under normal loads, and it is actually quite reasonable. In fact, since the machine is a little short on memory, there is often quite a bit of free CPU time floating around. (we are getting more memory soon) >- Replace the DZ by something better, probably some DH-11 clone. > These things use DMA for output, so you get far less interrupts. We would have done this quite a while ago, if it was necessary, but (again) the system does very little terminal i/o. Jeff Siegal - MIT EECS
ron@BRL.ARPA (Ron Natalie) (03/10/86)
Why don't you just bind whatever is bound to control s and q to other keys (^^ and ^\ seem poplular) or use a decent emacs that can handle flow metering to VT100's without needing the DC1DC3 flow control. -Ron
gwyn@BRL.ARPA (VLD/VMB) (03/14/86)
I think Jeff's terminal probably sends DC3s whether he has prepared UNIX to deal with them or not..
JBS%DEEP-THOUGHT@eddie.mit.edu (03/14/86)
Received: from MC.LCS.MIT.EDU by DEEP-THOUGHT.MIT.EDU via Chaosnet; 13 Mar 86 19:22-EST Received: from BRL-SEM.ARPA by MC.LCS.MIT.EDU 13 Mar 86 19:22:31 EST Received: from brl-vmb.arpa by SEM.BRL.ARPA id a003783; 13 Mar 86 15:45 EST Date: Thu, 13 Mar 86 15:36:45 EST From: Doug Gwyn (VLD/VMB) <gwyn@BRL.ARPA> To: Ron Natalie <ron@BRL.ARPA> cc: unix-wizards@brl-sem.arpa Subject: Re: dz-11 on a VAX-11/750 at 19,200 baud. I think Jeff's terminal probably sends DC3s whether he has prepared UNIX to deal with them or not.. Nope. I turned it off. Jeff Sigal -------