From: Bill Stewart <stewarts@ix.netcom.com>
Date: Sat, 18 May 1996 21:06:54 +0800
To: cypherpunks@toad.com
Subject: Re: Past one terabit/second on fiber[PHONE GEEK TALK]
Message-ID: <199605170553.WAA22076@dfw-ix6.ix.netcom.com>
MIME-Version: 1.0
Content-Type: text/plain


>>> "Wow", I said.  Far faster than the 2.5 Gb/sec transmission that is 
>>> currently fairly standard for long-haul fiber trunks.
>>The ads say they are selling it - that doesn't mean shipping it... yet
>>at least.  (Note that my employer is a direct competitor of Lucent 
>>so I have a vested interest in setting the facts straight)

No comment :-)  And I certainly don't presume to speak for Lucent....

Traditional fiber-optic technology is a hybrid between electrical and optical
components - big hulking multiplexers feed a high-speed electrical signal to 
a laser, which sends optical pulses down the fiber.
Every N km, a regenerator reads the photons, decides if it saw a 0 or 1,
and feeds that as an electrical signal to an output laser.  Repeat as needed.
The means that if you want to upgrade the signal speed, you not only need to
replace the muxes and lasers, you need to replace all the regens.
That's more of a problem for long-distance companies than locals
(the common FT Series G 1.7 Gb system uses them every 40 km.)

The new optical amplifiers not only go farther (e.g. 120 km),
but they do everything optically instead of dropping down to electrical
so they support a wide range of data speeds.  I don't know that
they can support the 1Tb experimental stuff, but they work fine
for the 8-color x 2.5Gb Dense Wavelength Division Multiplexing
stuff AT&T will be using.  This means that the first time you
do an upgrade, you need to rip out a bunch of regens, splice around
2/3 of them, replace 1/3 with opamps, and optionally replace the 
equipment at the ends with even bigger hulkinger multiplexers.
Depending on capacity needs, you may not fire up all 8 colors at once.
The next time you want to upgrade the same route, replace the muxes
if you didn't, or fire up more colors, or upgrade to the new Year 2005 model.
And since you've got to pick _some_ framing technology, it might
as well be SONET, which lets you build self-healing rings if you want
(the FDDI-like configuration, which not everyone uses, burns half
the capacity on restoration circuits, more than some of the less-healable
SONET configurations or mux-based restoration like AT&T's FASTAR.)

AT&T has announced that they'll be pouring lots of capital into this
over the next few years, partly to keep up with demand, and partly
to deploy SONET rings for faster restoration.

>I was figuring they'd cut out silences...as well as echo-suppression 
>cutouts.   Do they still do this?  Given modern fiber's capacity, I wonder 
>if they bother.

Voice compression and silence suppression aren't done domestically
(at least by most carriers.)  Undersea cables still use this,
though I don't know how much it's done on the newer fiber cables.
Of course, people running private networks do whatever they want,
and for overseas connections, people are often willing to trade
lower-voice-quality compression for the cost savings, especially
if the PTT on the far end is expensive.

>>I don't know that any number fiber cable is "standard" but 
>>36-fiber cable is not unusual.  To find the capacity of a cable, you
>>have to cut the number of fibers in half (as you did) because
>>generally each fiber is used only for a single direction of traffic.
>>You then have to cut it in half again because phone companies have 
>>everything redundant.  

A fairly common configuration for FT Series G is 8 fiber pairs
(one fiber for each direction), with 7 in service and 1 protection pair
to recover from mux-card and regen failures and other single-fiber hits.
To restore whole-bundle hits (e.g. backhoe fade), some of the 7 pairs
are typically dedicated to restoration - though seldom half.
The restoration pairs are often used for short-time reserved applications
such as TV connections for sports events, videoconferencing, or other 
applications where someone needs a lot of bandwidth for a short time
and is willing to be pre-empted or blocked if there's a failure.

>There's also a lot of 'dark fiber' out there, right?  Fiber laid as 
>part of a cable but not activated, because it's not yet needed.  

That's _highly_ location and company dependent.  Long-haul connections
are likely to be used efficiently, because you can get economies of scale
and because growth will fill them up (e.g. across the Rockies).
Short-haul connections (e.g. around town) are more likely to have
dark fiber because the big costs are digging up streets and paying
for government officials\\\\\\\\ licenses and permits rather than
the costs of the fiber you're putting in a trench once you dig it.
And the short-haul doesn't need regens, and often uses lower bandwidth
than the fiber can handle (OC3 muxes are much cheaper than OC48,
though per-Mbps they cost more.)

And of course, _everybody_ seems to want a T3 to the Bay Area or
other Internet-heavy locations.

>> I would add that much of a 
>>phone companies cost is in billing and customer service, etc.
>>Not the cost of installing and maintaining the fiber and equipement.
>
>This suggests that there would be a market for a LD phone company that 
>charged, say, a yearly payment of $200-300 for essentially unlimited use.
(The main 
>impediment to this would be regulatory; as I understand it LD companies have 
>to pay local telco's for connections on a per-minute connect basis.  Is that 
>right?  This needs to get fixed.)   Their billing costs would be very small. 

There's a lot of cost in switching equipment as well.  A feature-rich
voice telecom switch costs _far_ more per 64kbps voice circuit than the
1/(24*28*36=24192) fraction of a fiber that carries it.  And big muxes,
while cheaper than voice switches, are still expensive.

>>Internet telephony should make the use of bandwidth even more
>>efficient - thereby cutting costs.  The big guys who own
>>the fibers will still make money - the pipes that carry internet
>>traffic are still needed.  But the little guys will get squeezed out.
>>(until they become ISPs ;-).  Internet traffic could theoretically
>>be carried over this large amount of protection fiber (mentioned above)
>>that is out there for a much lower marginal cost than the current 
>>T3 or OC3 pipes that are being used. 

In addition to Internet telephony, ATM switch makers are doing voice
compression, and some of the fiber vendors are starting to use their
equipment to offer business voice services.  And voice-over-frame-relay,
which has more delay and therefore doesn't handle voice as well,
also is getting some market, especially internationally.

Pricing is a really difficult problem - if you price bandwidth proportional
to the 64kbps voice circuits, it's either too expensive for most businesses
to buy much or priced too low to make money on switched voice.
But if you price high-speed circuit bandwidth much cheaper than proportional,
it becomes cost-effective to buy customer-premises equipment and bulk bits
and run your own phone services.  Phone companies have been worrying about
this for video services for years, but fortunately Moore's Law and research
have let compression improve enough that you can run cheap video on
2*64kbps and good video on 6*64kbps, so they haven't been killed.
But Internet and similar data needs are starting to demand high bandwidth
at low costs, and the market will have to catch up somehow.
#					Thanks;  Bill
# Bill Stewart, stewarts@ix.netcom.com, +1-415-442-2215
# goodtimes signature virus innoculation