uccjcm@ecsvax.UUCP (John McLendon) (02/22/89)
We have an application where we need to essentially take over full control of the hp 9000/360 for up to 100 millisecs. Specifically, we need to detect the edges of a wave we read in real-time and take action based on the edge detection. The wave can last for up to 100 millsecs in time. What I'd like to know is if this is possible under hp-ux 6.21? We have found that the timer tics on the basic workstation were interfering with our measurement (causing us to miss an edge event), So we simply masked off interrupts. Can this be done under hp-ux? What are the possible problems? Will this cause the system to crash? Will this cause the system to lose clock tics? Help? Anyone? John... -- Signed: John McLendon uunet\ (919) 846-7931 (home) >mcnc!ecsvax!uccjcm (919) 941-5730 (play) gatech/
raveling@vaxb.isi.edu (Paul Raveling) (02/23/89)
In article <6534@ecsvax.UUCP> uccjcm@ecsvax.UUCP (John McLendon) writes: > >We have an application where we need to essentially take over full control >of the hp 9000/360 for up to 100 millisecs. Caramba! That's a whale of a long time -- When we designed EPOS we specified an absolute maximum interrupt latency (~time with interrupts disabled) of 1 millisecond on a PDP-11/45 because various devices couldn't tolerate anything longer. Also, a 1 millisecond latency needed to be an infrequent case. Typical measured interrupt latency was a lot less -- we didn't have hardware that allowed accumulating a mean latency measurement, but samples suggested it would probably be between 20 & 50 microseconds when the system was busy, negligible when the system wasn't busy. ---------------- Paul Raveling Raveling@isi.edu
rml@hpfcdc.HP.COM (Bob Lenk) (03/01/89)
> We have an application where we need to essentially take over full control > of the hp 9000/360 for up to 100 millisecs.... > So we simply masked off interrupts. Can this be done under hp-ux? > What are the possible problems? Will this cause the system to crash? > Will this cause the system to lose clock tics? Help? Anyone? Interrupts can only be masked by writing a custom driver. Doing so for 100ms could have nasty effects on other I/O that might be going on (you could certainly loose data on an ubuffered RS-232 card; I don't know the worst possible scenario). The system should be able to keep the system clock up to date. I'm not sure what the actual response time requirement is (eg. do you need to complete Xms of processing before the next edge, which could come as soon as Yms later?), or what else might be running on the machine at the same time. It's possible that a user process with real-time priority and locked in memory could handle the job without turning off interrupts. Again, I'm not sure what Basic workstation couldn't keep up without disabling interrupts; if it was also a 360 then HP-UX probably can't do any better, but if it was a 310 or a series 200 the faster processor might be all you need. Bob Lenk hplabs!hpfcla!rml rml%hpfcla@hplabs.hp.com