seb022@tijc02.UUCP (Scott Bemis ) (04/22/87)
From: Scott Bemis mcnc!rti-sel!tijc02 Texas Instruments, Johnson City, Tennessee Phone: (615) 461-2959 I am currently running Uniq Digital's UNIX V Release 2.0 Version 2 on a VAX 8600. I would like to know the lowest delay settings I can set on my UDA50 disk controllers. According to my DEC Field Service rep., there are three delay settings for UDA50 controllers. They are 10 milliseconds, 6.2 milliseconds, and 0 milliseconds. I would like to know if any one has had any problems using the 0 milliseconds delay on the UDA50 disk controllers.
gordon@prls.UUCP (04/23/87)
In article <147@tijc02.UUCP> seb022@tijc02.UUCP (Scott Bemis ) writes: > > >From: Scott Bemis mcnc!rti-sel!tijc02 ( several lines deleted - gpv) > According to my DEC Field Service rep., there are three delay > settings for UDA50 controllers. They are 10 milliseconds, 6.2 > milliseconds, and 0 milliseconds. I would like to know if any > one has had any problems using the 0 milliseconds delay on the > UDA50 disk controllers. I'd like to know what the trade offs are i.e. what considerations should be made in decideing which delay is appropriate. Thank-you, Gordon P. Vickers, (408) 991-5370, ======================= Signetics Corp. || Ultrix-32 ver 1.2 || PO Box 3409 M/S 69 || VAX 11/750 || Sunnyvale, California, USA 94086 ======================= {allegra,decvax,ihnp4,pyramid}!philabs!prls!gordon (WAS: prls!ems) Disclaimer: I reserve the right to disown any of the above text.
mangler@cit-vax.Caltech.Edu (System Mangler) (04/26/87)
In article <147@tijc02.UUCP>, seb022@tijc02.UUCP (Scott Bemis) writes: > I would like to know the lowest delay settings > I can set on my UDA50 disk controllers. Some DEC Unibus disks, such as the RK07 and RL02, have buffers too small to hold an entire sector. They suffer data late errors if they don't get at least half of the Unibus bandwidth. Since they transfer only one word when they get the bus, they can be easily starved by devices (like the UDA50) that transfer many words per bus arbitration, even if the RK07/RL02 is at the head of the bus. Thus, the purpose of the Unibus Delay jumper is to leave lots of idle bus time for such devices. The best setting is usually determined by experiment. Try the fastest setting, and if you get data lates, try a slower setting. With some combinations of devices, no setting is slow enough. (We solved that one by selling both the RL02's and the UDA50's). I think DEC always sets them to the slowest setting. The corresponding DMA speeds are approximately: 0us 800 KBytes/sec (same as an RP04, circa 1975) 6.2us 350 KBytes/sec 10us 250 KBytes/sec (half the speed of ST506) Don Speck speck@vlsi.caltech.edu {seismo,rutgers,ames}!cit-vax!speck
chris@mimsy.UUCP (Chris Torek) (04/26/87)
>In article <147@tijc02.UUCP>, seb022@tijc02.UUCP (Scott Bemis) writes: >>I would like to know the lowest delay settings >>I can set on my UDA50 disk controllers. In article <2477@cit-vax.Caltech.Edu> mangler@cit-vax.Caltech.Edu (System Mangler) writes: >Some DEC Unibus disks, such as the RK07 and RL02, have buffers too >small to hold an entire sector. They suffer data late errors if >they don't get at least half of the Unibus bandwidth. ... >Thus, the purpose of the Unibus Delay jumper is to leave lots of >idle bus time for such devices. >... With some combinations of devices, no setting is slow enough. >(We solved that one by selling both the RL02's and the UDA50's). The 4BSD Unibus allocation code understands `bus eater' devices like RK07s, and will lock all other devices out of the Unibus while the bus hog is using it. I believe the RL02 driver is not marked as an exclusive-Unibus-use device, but that is a trivial change. The problem is that the stock 4BSD UDA50 drivers ignore the bus lock out protocol. If you run 4.3BSD and suffer with UDA50s and RK07s and/or RL02s, you can install the driver I posted to net.sources a few weeks ago, or (for ARPA sites) FTPable from host mimsy.umd.edu as uda50/4.3.shar. This driver honors the bus lock protocol, and you can set the delay down to zero. > 0us 800 KBytes/sec (same as an RP04, circa 1975) >Don Speck speck@vlsi.caltech.edu {seismo,rutgers,ames}!cit-vax!speck I only get 750KB/s. Still, that is not too awful. Alas, actual throughput from any one drive never seems to rise above 125KB/s, because of the delays for seeks and head switches. Just think: another 13 RW amplifiers per RA81, and actual throughput would probably double. But DEC must have saved perhaps $50 per drive.... -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7690) UUCP: seismo!mimsy!chris ARPA/CSNet: chris@mimsy.umd.edu