buck@siswat.UUCP (A. Lester Buck) (04/27/89)
[I am redirecting followups to comp.sys.ibm.pc.rt, which is where this stuff belongs, and the rt group could use more traffic ;-)] In article <643@jc3b21.UUCP>, crash@jc3b21.UUCP (Frank J. Edwards) writes: > > Have you noticed that access to the floppy drive doesn't slow down > the machines processing at all? whereas on other boxes, reading or > writing from/to the floppy can practically stall the box... Yeow! How could it slow down any? It is already the SLOWEST floppy drive in the known universe. Go for a coffee break while it formats a 1.2M disk. Yes, it does do verification and this can be turned off, but still it works like molasses. Backing up VRM (the cvid command) onto two disks takes something over twenty minutes. And after any use of the floppy drive, our system logfiles are peppered with hardware error entries from the floppy drive. (No bad spots on disk, just bunches of retries or whatever.) > I haven't coded any AIX device drivers, but I have studied them > somewhat, and my conclusion is: I would prefer to write AIX device > drivers that some other kernel's. Of course, you still the *real* > hardware device driver which is part of the VRM :-) Exactly - instead of writing one piece of code, you get to write two pieces, and support another interface. One great advantage of VRM is the ability to dynamically add a driver to a running Unix system without rebooting. The only other version of Unix I have heard with this feature was the Unix PC. A disadvantage of VRM drivers is that, with no AIX source available, if the supplied drivers for ESDI or SCSI don't do what you want, or have bugs, you are basically out of luck. [Take the SCSI adapter implementation on the RT - so brain-dead that any working implementation is a miracle. For example, the hardware has separate operations for moving data from the SCSI bus onto the card and moving the data from the card into system memory. Maybe it just never occurred to them to use a FIFO, but the driver then gets to manage 128 512-byte buffers. Sounds fun, huh? And the VRM driver started out presenting a SCSI virtual disk, then was hacked over until it sorta has a passthrough mode to handle the dozen other types of SCSI devices, as long as you don't want to transfer more than 6K bytes at one time. I was told by the current VRM SCSI driver maintainer "we have learned a lot from the SCSI adapter". Local IBMers say that the MicroChannel SCSI adapter coming with the new RT will be "the card of the year", whatever that means. Sure...] -- A. Lester Buck ...!texbell!moray!siswat!buck