Subject: Re: The demise of DEV_BSIZE
To: Ignatios Souvatzis <is@jocelyn.rhein.de>
From: Bill Sommerfeld <sommerfeld@orchard.arlington.ma.us>
List: tech-kern
Date: 10/07/1999 17:38:13
> Now, with the 1536 byte DomainOS sectors somebody else mentioned: this might
> prove useful.

The lore I picked up by osmosis while working at apollo:

I find the 1536 number difficult to believe.  1036, i can believe --
it would allow for 1024 bytes of data, an 8-byte aegis UID, and 4
bytes for "something else".

I believe that disks on apollos were typically formatted to store 1K
of data (the original VM page size of early apollo 68k boxes) plus a
small amount of metadata (which I think included the UID of the file
which the page was part of).  If the UID in the disk page didn't match
the uid of the file it was supposed to be a part of, the
kernel/filesystem got appropriately unhappy.

As an aside, aegis played some interesting games so that the token
ring driver would PIO in the packet header into a small buffer and
then start DMA in the data part into an aligned data page which could
then be page-flipped into the address space of whichever user process
had mapped in the file..

					- Bill