mp@allegra.UUCP (Mark Plotnick) (01/08/87)
After a recent session of unibus card musical chairs on our 4.2BSD 780 (which involved removing a dr11w and one of two 3com ethernet boards and replacing an lp11-like device with a different one), we noticed that system cpu time increased a lot whenever there was sustained output on our dz's. Simply catting files on 2 9600 baud lines simultaneously, for instance, would increase system cpu time to about 50% and make the interrupt rate as reported by vmstat go up to about 1800. (about 1 interrupt per character output). We didn't think this was normal, since 4.2 is supposed to use some pseudo-dma method to cut down the interrupt rate. We don't have any other 780 running 4.2, but on a 4.2 750 (with 1 dz and 1 3com board on the unibus), vmstat shows about 5 interrupts per second when doing 9600 baud output on a line (the 780 would show around 1000). DEC's diagnostics indicated the uba was bad, but the FE came to the conclusion that this was due to the remaining 3com board, so we put in an interlan, but this didn't reduce the interrupt rate. The FE thinks all the NPG wires are correct, and pulling the ethernet and lp boards out doesn't change things. What are we doing (or thinking) wrong? Here are the devices on the 780: uba0 at tr3 dz0 at uba0 csr 160100 vec 300, ipl 15 dz1 at uba0 csr 160110 vec 310, ipl 15 dz2 at uba0 csr 160120 vec 320, ipl 15 dz3 at uba0 csr 160130 vec 330, ipl 15 dz4 at uba0 csr 160140 vec 340, ipl 15 dz5 at uba0 csr 160150 vec 350, ipl 15 dz6 at uba0 csr 160160 vec 360, ipl 15 dz7 at uba0 csr 160170 vec 370, ipl 15 il0 at uba0 csr 164000 vec 400, ipl 15 lp0 at uba0 csr 177514 vec 200, ipl 14 dkitdr0 at uba0 csr 167770 vec 420, ipl 15 [ dr11c for datakit ] (the massbusses are occupied by 2 emulex Sc7000's, a TU-78, and an RH780, if that matters). Mark Plotnick allegra!mp