[comp.arch] Why optical disks are slow to seek etc...

die@frog.UUCP (Dave Emery, Software) (11/10/86)

In article <586@unccvax.UUCP> jow@unccvax.UUCP (Jim Wiley) writes:

>Optical Storage International has a WORM optical disk with a transfer
>rate of 262K bytes per second.  Thats about half as fast as a ST506
>winchester.  The killer on the data rate is the 1.33M bits per
>second SCSI interface.  I admit the seek time is poor, but SCSI
>seems to me to be a pretty brain damaged way to sacrifice performance
>for flexibility.


	I am afraid to say that the transfer rate through a SCSI bus
with efficient interfaces on both ends is something more on the order
of 1.5 megaBYTES per second rather than the 1.33 megaBITS per second
cited. 

	 SCSI timing in the slower asynchronous mode that most people have so
far implemented is interlocked but both logically and electrically 
asynchronous <no bus clock> so the speed at which *both* ends respond to
each other determines the effective transfer rate. The last time I looked up
the timing in the SCSI standard (which keeps changing in small and subtle
ways) and did the throughput calculation the maximum rate was about 1.5 mbytes
sec (it is remotely possible, however, that the current standard SCSI timing
actually maxes out at 1.33 megaBYTES/sec).

	Charles River's Versabus SCSI interface, designed in 1981 when
the SASI/SCSI standard first came out, (and thus a discreet MSI TLL design
rather than one using the new VLSI SCSI chips such as NCR's or Fujitsu's)
has been measured at around 1.1 - 1.25 megabytes/second with certain real
controllers that have fast interface logic.  But ST-506 Winchesters (which
use a 5 mhz data rate) cannot possibly source or sink data faster than
833 kbytes/sec. With normal formating including gaps, preambles, address
and data marks, and sector headers this is reduced to about 550 kbytes/second
or about half of the SCSI bus bandwidth obtainable with straightforward 
interface logic.  Thus WORM drives could be at least twice as fast
as ST-506 devices before they were significantly limited by the SCSI bus.

	The SCSI standard defines a synchronous mode with a transfer rate
of 4 mebabytes per second.  When this mode actually comes into use
and is supported in both host adapters and controllers, it will be possible
to connect WORM optical drives that transfer data about 6 or 8 times faster
than ST-506 drives to a SCSI bus without performance being limited by
the bus transfer rate.

-- 
          David I. Emery   Charles River Data Systems
983 Concord St., Framingham, MA 01701 (617) 626-1102 uucp: decvax!frog!die

stuart@bms-at.UUCP (Stuart D. Gathman) (11/11/86)

In article <586@unccvax.UUCP>, jow@unccvax.UUCP (Jim Wiley) writes:
> winchester.  The killer on the data rate is the 1.33M bits per
						  ^^^^^^^^^^
> second SCSI interface.  I admit the seek time is poor, but SCSI
> seems to me to be a pretty brain damaged way to sacrifice performance

I beg your pardon?

Perhaps you meant 1.33M *bytes* per second.  SCSI is 2 to 4 times faster
than ST506.  ST506 is limited by definition to 5M bits per second.  SCSI
goes 10M to 20M bits per second.  The 20M bps is as good as a high
performance mini.
-- 
Stuart D. Gathman	<..!seismo!{vrdxhq|dgis}!BMS-AT!stuart>