Re: devfs

Bob Tracy (rct@gherkin.sa.wlk.com)
Wed, 14 Jan 1998 07:29:40 -0600 (CST)


rgooch@atnf.CSIRO.AU said:
> I recall someone telling me that Solaris "stripes" are not actually
> partitions. Their strips fit inside a "real" partition. Is that
> correct? I don't really find the "s" clear.

Time to lay it to rest, but here's a way for folks to live with 'd'
and 's'. Think 'd == device, s == subdevice' and assign whatever
meaning is consistent for a particular type of device. In the
System III/V way of naming things, 'd' was never intended to represent
'disk'. The original meaning of 's' is a bit more obscure, and 'slice'
was at least as common an interpretation as 'subdevice' as I remember
it. Regardless, 'stripe' or 'strip' wasn't one of the valid meanings.
In my mind at least, there's a natural correlation between a
subdevice and a partition, and 's' isn't as limited as 'p' if/when
you want to apply the naming scheme to a device that doesn't have
partitions.

An example: for SCSI devices, '(d)evice == LUN'.

One more example: ESDI disks. On the old 3B2 computers with ESDI
disks, a SCSI target ID (t1-t7, t0 was reserved for the host adapter)
corresponded to a SCSI <--> ESDI controller (Everex) that could
support up to four spindles (d0-d3). Simple math implies you could
hang up to 28 spindles off a single SCSI host adapter if you were
crazy enough to put up with the I/O bottleneck.

The ESDI example in particular is why I would prefer 'd' to 'u', but
I can live with 'u' if that's the consensus.

-- 
Bob Tracy               |  If you have any trouble sounding condescending,
Firewall Security Corp. |  find a Unix user to show you how it's done.
rct@frus.com            |  	-- Scott Adams: DNRC Newsletter 3.0