🚀 go-pugleaf

In article <94@zds-ux.UUCP> gerry@zds-ux.UUCP (Gerry Gleason) writes:

+---------------
| That reminds me, I've been thinking about the possibility of fiber as a
| replacement for SCSI and/or present LAN technologies.  The application
| would be more typical desktop micro's, not killer micro's so the bandwidth
| doesn't need to be pushing the limits, and cost is a much more important...
| 	2)  Bandwidth fast enough to beat (or at least tie) what it
| 	    replaces (SCSI = 1-40MB/s =~10-300M bits/s >> Ethernet).
| So, is this reasonable?  What are the limits of "cheap" opto-electronics
| (for me, cheap is < 10$ for a transmitter and reciever...
+---------------

Well, plastic fibers (e.g. H-P 1mm plastic "snap-in" links) can be made
to run at Ethernet speeds at distances to 25 meters; I've done it. (Yes,
they're only rated for 5 Mbaud, but if you use "impulse coding" you can
get 10 Mbit/sec easily and *very* reliably. And impulse coding can be
converted/deconverted to Manchester, so you can use standard Ethernet
chips.) And an active multi-port star coupler is very simple, and needs
no clocks, due to the impulse coding.

But much above Ethernet speeds, the pulse dispersion in the fiber kills
you. 1mm fiber has a numerical aperture of about 0.6, which limits the
distance to not much more than a bit-time's "length". For example,
10 Mb/s -> 100ns -> ~20 meters. It's better than that (obviously, see above),
but not much. And unfortunately, the cost of a pair of transmitter/receivers
plus several meters of plastic is not much cheaper than today's Ethernet
transceiver chips plus some RG-58C/U. It's a technology whose time passed
it by. (*sigh*)   (...unless you *have* to have very high isolation.)

In any case, glass fibers aren't that expensive anymore, and plastic-
packaged optics for glass have appeared which will handle plain 20Mbaud
data (and therefore 10 Mb/s Manchester), so at 10Mb/s I'd just use some
kind of fiber-"Ethernet"-with-active-hub. Rumor has it that the standards
folk will look at 10baseF right after 10baseT gets steadied down.

The next convenient speed above that is 100 Mb/s, which you can get up to
several hundred meters with relatively cheap glass or PCS fiber, plastic
package 900nm optics, and AMD "TAXI" serializers. Cost per end of a full
duplex "fiber UART" (optics + TAXIs) is less than $100, and likely to fall.
This is FDDI speed (uses FDDI's 4b/5b coding), but *much* cheaper. With a
simple "byte bridge" active hub, you could use it like a 100Mb/s "Ethernet".
The hub would cost just a few bucks + <$100 per port... about the same or
less than a 10baseT hub.  I've nicknamed it "100baseF", but I doubt it will
really get standardized any time soon... ;-}  (FDDI is too "hot" right now.)

And then there are the gigabit speed devices I mentioned in the previous
article. "1000baseF"??? Well, for now gigabit/sec optics are *very* expensive
(need lasers?), but cheap coax works for ~30 meters if you can accept wire.

So that's the range of links I know of. For the "fiber-as-I/O-bus" I'd
probably choose the 100Mb/s speed, since it could handle even fast SCSI.
(Note that it's also roughly the same speed as a fast IBM 370 I/O channel.)
For performance, you'd need a link per device; for cost, you'd share a
"channel" among several devices with a hub.

-Rob


-----
Rob Warnock, MS-9U/510		rpw3@sgi.com		rpw3@pei.com
Silicon Graphics, Inc.		(415)335-1673		Protocol Engines, Inc.
2011 N. Shoreline Blvd.
Mountain View, CA  94039-7311