9 Dec
2003
9 Dec
'03
12:30 p.m.
I wrote:
> If the drive does not use buffered seek, it is
necessary to transfer
> a new cylinder's data into the buffer RAM in less than the seek time.
> As far as I know, the highest capacity 5.25-inch Winchester drives that
> did not use buffered seek had no more than four heads, 615 cylinders,
> and no less than 20 ms seek time, so the maximum transfer rate needed
> to transfer data to and from the backing disk is about 33 MB/s. This
> would require RAM with a 30 ns cycle time. More practically, the
> buffer RAM could have a x16 or x32 organization, stretching the
> cycle time requirement to 60 ns or 120 ns.
"James Dickens" <jdickens(a)ameritech.net> wrote:
okay not an expert, but most systems using these
drives used interleaving
of
sectors(because there was no way the system was fast enough to handle the
data), the IBM PC used a factor of 6.
Interleaving by the host is completely irrelevant to the disk emulator,
which doesn't even have a concept of "sectors". To the disk emulator,
a track is just a big collection of samples of the data line, which
don't even have a 1:1 correspondence with host data bits (or even host
channel code bits).
The machine the device would not
know
what to do with 30MB/s of data if you produce it.
The 33 MB/s I quoted was the necessary transfer rate between the buffer
memory and the drive *inside* the disk emulator. It has nothing to
do with the host computer.