RALPH@UHHEPG.BITNET (02/07/88)
Date: 6-FEB-1988 14:45:29.94 From: Ralph Becker-Szendy RALPH AT UHHEPG To: 0::"info-vax@kl.sri.com",RALPH Subj: Re: Wild processes Consuming CPU > Date: Tue, 5 Jan 88 11:44:00 CDT > From: Z919016@UTMBEACH > Subject: Wild Processes Consuming CPU > > Has this happened to any else? We have a VAX 750 running > VMS 4.5. Twice in the last 3 months, a process has run > wild and consumed hours of CPU time until noticed and > killed (this occurs during low-use hours). > > First process was in FOCUS, a database package. Second > process was in MAIL, and in addition to 2 hours 28 minutes > of CPU time used in 2 1/2 hours, had a Buffered I/O count > of over one million. In both cases, the users had > gone home, and hadn't done anything unusual. Yes, on a VT100 connected via 500 ft of cable to a 780 using all modem control lines. The user switches his terminal off without logging off (staying in DCL). Then the process starts using around 6% of the available CPU time (absolutely constant, not just on the average) and creating several IOs per second. As the process is using CPU time, it is not going to be automatically terminated by the WATCHER program. The funny part is: there is no image being executed, it is just DCL; but the CPU time is being used in USER mode, not on the interrupt-stack, or in kernel mode. To cure the problem, i just switch the terminal back on, it immediately stops using CPU time, and it is still in DCL. Also, if the user logs out BEFORE switching the terminal off, nothing funny happens. All this happens only with this one terminal, using this one cable, but on any port with modem control lines. We don't really understand the problem, but i have the following theory: When the terminal goes off, its V24 line drivers are not actively driving the modem control lines any more, so they pick up electrical noise. Actually, we can see a lot of AC noise on all the output lines of this terminal (around 2V p-p). Maybe the input lines on the VAX port are too sensitive, and actually register the noise as being modem control lines changing their state at around 60 Hz, and this drives the terminal driver crazy. I just don't understand how logging off could cure this problem. But probably my theory is completely wrong. Ralph Becker-Szendy RALPH@UHHEPG.BITNET University of Hawaii / High Energy Physics Group (808)948-7391 Watanabe Hall #203, 2505 Correa Road, Honolulu, HI 96822 "Hawaii - it's not just for tourists. People actually live and work there."
carl@CITHEX.CALTECH.EDU (Carl J Lydick) (02/08/88)
> Yes, on a VT100 connected via 500 ft of cable to a 780 using all modem control > lines. The user switches his terminal off without logging off (staying in DCL). > Then the process starts using around 6% of the available CPU time (absolutely > constant, not just on the average) and creating several IOs per second. As the > process is using CPU time, it is not going to be automatically terminated by > the WATCHER program. > > The funny part is: there is no image being executed, it is just DCL; but the > CPU time is being used in USER mode, not on the interrupt-stack, or in kernel > mode. To cure the problem, i just switch the terminal back on, it immediately > stops using CPU time, and it is still in DCL. Also, if the user logs out BEFORE > switching the terminal off, nothing funny happens. > > All this happens only with this one terminal, using this one cable, but > on any port with modem control lines. > > We don't really understand the problem, but i have the following theory: When > the terminal goes off, its V24 line drivers are not actively driving the modem > control lines any more, so they pick up electrical noise. Actually, we can see > a lot of AC noise on all the output lines of this terminal (around 2V p-p). > Maybe the input lines on the VAX port are too sensitive, and actually register > the noise as being modem control lines changing their state at around 60 Hz, > and this drives the terminal driver crazy. I just don't understand how logging > off could cure this problem. But probably my theory is completely wrong. We've had a similar problem here. First: Note that your cable doesn't conform to the specs for RS232 communications (it's about 10 times longer than the specs allow), but since most RS232 interfaces are overdesigned, such cables are not generally a problem. In our case, the problem was with a 3-conductor cable to an H19 terminal in the basement (our computer's on the 2nd floor). The user owning the terminal tended to turn off the terminal without logging out. The result was that the (now unterminated) cable seemed to have enough coupling between the input and output lines that any character sent to the terminal resulted in some character being sent to the VAX. This meant that when a job was still logged in, we got LOTS (your 6% CPU usage sounds a little low) of chatter on that line. The reason, in our case, was that, with a process logged in, everything apparently sent from the terminal got echoed back, resulting in further noise on the line, which caused more echoing from the VAX, etc. Without a job logged in on the line, there was no echoing (actually, a control-y or control-m would have started a loginout, but those characters were NOT generally generated by noise on the line). We had a similar problem on a line from somebody's PC to a laser printer. You might try putting a resistor between signal ground and the transmit data pin on the terminal as a solution if you can't get the user to log out before turning off the terminal.
rrk@byuvax.bitnet (02/10/88)
We have the same thing happen just about any time someone's process becomes disconnected from the physical terminal. We'be had it happen in ALLIN1 and KERMIT and DCL I think.
rrk@byuvax.bitnet (02/10/88)
I should have made my other message more complete. I said it happens frequently to us when a process's virtual terminal becomes separated from the physical one. We operate on LAT (decserver 200) ports connecting to virtual terminals.