[Simh] Disk info request

[Clem Cole]

On Wed, Mar 9, 2016 at 10:46 AM, Mark Pizzolato <Mark at infocomm.com> wrote:

> I explicitly recall on Ultrix contriving entries in
> /etc/disktab to describe arbitrarily sized disks connected to 3rd party
> MSCP controllers.  The info in /etc/disktab is used by the file system
> creation code to lay down the file system data structures scattered
> around the disk surface based on some relationship of the presumed
> geometry that was described.
>

​right -- when Leffler wrote "newfs" which was the front end replacement
for mkfs and bunch of other utilities, disks were still "geometry
oriented"​ and a lot of work in FFT (aka UFS) had gone into trying let the
kernel know about the actual geometry being used, so better seek scheduling
could be done.    At the time there was ton of research being done in the
area (if you look at the USENIX papers - you find a couple including one or
two with my name of them).

Anyway, a few things changed... local caches were introduced on the drive,
which mitigated a lot of what UNIX was doing because they did their own arm
scheduling.   Plus drive manufacturing got better at packing sectors, so
when before the max number of sec/cyl was a fixed amount, the larger drives
now sec/cyl changed depending on the drive.    Much less the whole bad
block stuff would screw up all the OS work. Finally,  SCSI introduce the
idea of a single Logical disk block (i.e. the disk was just a linear array
of blocks) and all the scheduling was left to the controller microcode
(i.e. a dumb OS was better than one that tried to out smart the controller).

Clem
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.trailing-edge.com/pipermail/simh/attachments/20160309/d2914e2c/attachment-0001.html>