UFS filesystems on SCSI disks

[der Mouse]

In article <1310 at creatures.cs.vt.edu>, alanlb at csgrad.cs.vt.edu writes:
> I'm trying to create UFS file systems on a SCSI disk and have been
> wondering what to specify for the number of tracks/cylinder.

The filesystem layout code wants to know two numbers: the number of
sectors per track and the number of tracks per cylinder.  (The number
of cylinders is implicit; it can be calculated from those two numbers
and the total size of the filesystem.)

Sectors per track I assume you can deal with.  Tracks per cylinder is
the number of recording surfaces available, also known as the number of
heads.  (Of course, you want to specify only the number of heads
available for user data storage.  If the drive or controller reserves
one or two surfaces for bad block replacement, you should ignore
it/them.)

> Logically, my SCSI disk has 1284 tracks and 1 head (Seagate 1280N).

Are you sure of that?  I didn't think anyone made single-platter disks
any longer (well, except for removables).  Unfortunately I don't have
any reference in which I could look up Seagate drive numbers, so I
can't check that myself.

> Should I specify 1 track/cylinder,

If you're really certain there's only one head, yes.  But that seems
rather unlikely to me.

> (BTW, I told rdb the disk has 64 sectors/track)?

Again, is this actually true?

Of course, there's another side to this.  Since the filesystem wants
this information to attempt to optimize block placement and such, it is
substantially less important with modern disks, where the disk can
often overrun the bit pipe between the disk and main memory, where
there's a large buffer on the disk or controller card, where the
controller already does read-ahead, etc....  In any case, the only
effects of giving incorrect numbers will be (a) efficiency impairment,
because the filesystem's idea of optimal block placement won't really
be optimal and (b) slight loss/gain of storage space due to changing
filesystem overhead.

					der Mouse

			old: mcgill-vision!mouse
			new: mouse at larry.mcrcim.mcgill.edu