stevens@hsi.UUCP (Richard Stevens) (08/20/86)
The UDA50 driver supplied with 4.3 sets the "burst rate" of the
driver to 3 (UDABURST - 1, where UDABURST is #defined to 4 at the
top of the driver). Not having any documentation at all on the
UDA50, I'm not quite sure what the burst rate is, but I'd guess
its the number of transfers on the Unibus before the controller
gives up the bus to someone else ?? Anyway, the 4.2 uda.c that
we were running (from Bob Brown at RIACS, Revision 2.1, dated
84/03/05) in effect set the burst rate to 0 - it never set it
explicitly, but the appropriate bits were defaulted to zero.
With the burst rate set at 3, we were getting a lot of bus-grant
lates from an Emulex TC11 tape controller driving two Kennedy 9300
tape drives (125 ips). Only when we set UDABURST to 1 (setting the
burst rate to 0) did the problems subside. If you have any devices
like the TC11 (8 bytes of buffer!!) on a busy Unibus, you should probably
lower the burst rate in the driver.
Richard Stevens
Health Systems International, New Haven, CT
ihnp4 ! hsi ! stevensdavest@tektronix.UUCP (Dave Stewart) (08/25/86)
It is fairly standard practice in these parts to put a UDA-50 on its own unibus. DEC recommends putting a UDA-50 only with other DEC hardware on a unibus, but we have found that even this can cause problems. An interesting development of this is that the VAX 8650 we have here at Tek has a unibus that is designed to have only a UDA-50 because there are no slots for anything else! (Actually, even this can be bent since we got DEC to squeeze a DR11-B on this bus because of timing problems with another device). This strikes me as humorous: most peripheral manufacturers design their equipment around CPU features. Here poor DEC has to design their CPU around their peripherals! Anyway, I degress. If you have more than one unibus, the good money is on moving everything off of the bus that has the UDA-50. Then you should not need to worry about the burst rate or other device-bus contention problems. My apologies to DEC if I have oversimplified this or overstated it. -- David C. Stewart uucp: tektronix!davest Unix Systems Support Group csnet: davest@TEKTRONIX Tektronix, Inc. phone: (503) 627-5418