stevens@hsi.UUCP (07/02/84)
There have been some questions on the net concerning the DMA versus programmed i/o mode of the asynchronous ports of the DMF32, plus I have been wondering what the DMF buys you over the DH, so I ran the following timing tests. The tests consisted of continual output (cd /sys/vaxuba; cat *.c) to 7 terminals (CIT-101e's) on an 11/750 running 4.2. I monitored this on the console with a "vmstat +5". This is where the %sys column came from. I averaged 20 of the vmstat's numbers together - I don't know how accurate the vmstat us/sy/id percentages are? I then ran a smaller output test (cd /sys/vaxuba; time cat *.h) to see what the user, system and clock times were, again on 7 terminals simultaneously. That is where the system time and clock time columns came from (the 7 values, averaged). (All of the user times were 0.0.) "vmstat" "time cat *.h" mode %sys sys time (sec) clock time ------------------- ------- -------------- ---------- DEC DMF32, DMA 64.6 % 7.14 1:29 Emulex CS21/H (DH) 67.2 % 7.17 1:29 DEC DMF32, prog i/o 70.1 % 7.54 1:29 Note that 7 terminals, each with about 7-8 seconds of system time over 89 minutes, corresponds quite well to about 65% of the system (the vmstat number). From these numbers, the DMF in DMA mode is barely better than the DH, but the difference between prog. i/o and DMA isn't that great. Note that these tests (output only) aren't the best. I'd guess that a good DMF driver that can dynamically switch between programmed i/o and DMA modes would probably be the best for real world tty i/o, but it probably won't be that much better than the DH. If you're interested in some comparisons of the DH and the DZ, take a look at the "Emulex Controller Handbook" (free from Emulex) as they have some output only tests run under VMS. (This book has some better information on the DEC bus'es and controllers than DEC's Peripheral Handbooks.) Richard Stevens Health Systems International, New Haven, CT { decvax | hao | seismo | sdcsvax } ! noao ! hsi ! stevens ihnp4 ! hsi ! stevens