Subject: Re: Of SCSI disks and things amiss
To: None <tech-kern@NetBSD.ORG>
From: Chris Csanady <>
List: tech-kern
Date: 03/01/1996 00:44:33
>In many cases, the actual disk *does* extend beyond what the `geometry'
>implies; even after you discount reserved sectors.  If you make the
>change you suggest, this space would become unusable.
>It seems more appropriate to either:
>a) stop pretending we have any idea what the geometry of a SCSI disk 
>is, and just make it one large track, or

Ok, here is what I have to say, again.  I didnt send my previous
reply to the right place. :|

Anyway, I think this is something worth looking into further.  The
allocation policies in ffs (or any filsystem) really should not be
based on some fictitious geometry.  This seems like it could be
a place to reduce the complexity of the fs code, and increase performance
at the same time.  I think that is generally good. :)

Also, I would like to propose an idea.  I am not sure if it would work
on all disks, but I am curious.  If a disk reported the *actual* size,
would it not be possible tell which sectors were phsically in the same
cylinder?  ie, if you divided the total disk by the number of heads,
you could theoretically calculate the offset of each platter.  And if
a sectors were at any common offset from each platter, they would be
in the same cylinder.  so instead of abstracting the disk as a single track,
why not as a single cylinder.  I think this could be of considerable
use to the file system in block allocation. :)

Is this feasible?


>b) make the warning from disklabel(8) non-fatal.