🚀 go-pugleaf

In article <8167@lindy.Stanford.EDU> ralerche@lindy.Stanford.EDU (Robert A. Lerche) writes:
> A simple way to grok disk speed: 60 revolutions / second is the typical
> hard disk speed.  So, with 17 sectors on a track (again, typical), 512
> bytes/sector, we have around 500 K bytes/second (about 30 M bytes/minute)
> at 1:1 interleave.

Eg: this is *absolute* maximum thruput from a ST506 drive - no missed 
rotations, exactly one rotation per track, 1:1 interleave.  This is
rather hard to achieve (or utilize for that matter) in most cases....

Some benchmarks I've done suggest that typical random reads of 4-8K through
UNIX buffered disk devices (not files) is typically *less* than 1/4 of that.
A surprising number of "reasonable" machines get an effective bandwidth
of 25-50K/sec at most....

Micropolis 1570 (SCSI) or 1550 (ESDI) series, on the other hand, is

	512 (sectorsize) * 36 (sectors/track) * 3600 / 60:
	1105920 bytes/sec.

(Most other SCSI and ESDI drives should be the same, if not a bit more
depending on bit encoding)

So, roughly double the effective bandwidth - though, some disk drivers
are capable of doing considerably better.

>  ... So, using raw bitmaps it's a problem to drive a 100
> page/minute marking engine.  

You're not kidding.

> However, with some fairly simple compression it's probably good enough.

It depends on "how busy" the bitmap is.  If it's text and line drawing
a simple RLL encoding would work well, but rasterized images may often
be a different story...
 
> A historical note:  60 revs/second is the same speed that mainframe hard
> disks have run at for years and years.  Some of IBM's disks run a bit faster
> nowadays (I think the current 3380's go at about 80 revs/second).  The
> mainframe disks have much more data per track, though, and so achieve
> much higher transfer rates (up around 4 M bytes/second).

CDC Hydra goes at 120 revs/sec.

(US Export Control requires special licensing for drives that rotate
faster than X RPM, I'm not sure whether "X" is 3600 or slightly higher)
-- 
Chris Lewis, Elegant Communications Inc, {uunet!attcan,utzoo}!lsuc!eci386!clewis
Ferret mailing list: eci386!ferret-list, psroff mailing list: eci386!psroff-list