Post by Mike S Post by Paul Post by Mike S Post by Brian Gregory Post by J. P. Gilliver (John)
Post by Brian Gregory
A DSL modem that's switched on during daylight hours will often
have trouble continuing to work after dark. Switching it off and
on will get it going again.
Quite why they seem to be unable to make them so that they can
tell when the error rates are getting alarmingly high and
automatically recover on their own I don't know. But I've never
owned one that wasn't quite happy to just sit there for hours
reporting that it couldn't decode anything because there was too
much line noise for the speed it had initially negotiated when
conditions were better.
Even once a DSL modem has adjusted to the worst conditions
(usually night time) (by being switched off and on late at night)
it's likely it'll occasionally need switching off and on again,
maybe about once a week.
My cheap dynamode one, plugs away, day and night, very rarely
requiring any action on my part, and has for many years.
The ISP can limit the speed of the connection to make it more
reliable, or you can be so close to the exchange that you connect at
the max speed with lots of noise margin every time, or maybe you are
a long way from the exchange so that the troublesome higher
frequencies are virtually unused by the modem.
Exactly right, I worked for an ISP and we 'capped' (limited the
higher frequencies used by the connection) on problematic lines. In
our spare time we'd look for really slow connections (using a google
map that showed the connection speeds with different color markers
for speed ranges) and tried optimizing them for speed and stability.
That's a function of architecture though.
The old system used 18000 feet (or optionally 36000 feet) of
wire, to connect subscribers all the way back to the CO. On large
operations, the operator simply applied a "blanket cap" and didn't
give a crap. They took 8Mbit/sec max ADSL1 and sold a service
advertising 5Mbit/sec, and then capped it at 3Mbit/sec without
ever examining the statistics. They had the option of selling
it as 3Mbit/sec service, but they didn't, and... they got away
with it too.
The new system uses fiber-to-the-corner, the wire length (final hop)
is closer to 500 feet, as the wire runs from the box on the
corner of your street, to your house. And when they sell
you a service at "X", they actually deliver "X". Shurely
a miracle. No more cap, except for the cap of the
advertised service of X. No more laddling SNR margin
randomly and at their discretion, on top.
Some customers here, used to use DMT and file a trouble
ticket with the ISP, to "fix" the first case. And actually
have the link adjusted properly. Some of those people,
hanging out at DSLReport :-)
There are still areas of the country operating the old
way. And the operator in that case, has absolutely no plan
to fix any infrastructure. It'll take a slap from the government
to keep the physical plant functional. There's a guy in the WinXP
group who is getting the old fashioned "service", complete
with "horse, buggy, and excuses".
We're getting fiber installed (Santa Cruz, CA) in the city center areas
now. I stopped using DSL because where I live, even though I'm less than
.75 mile from the CO and got great DSL speeds, the phone wiring is so
old that when it rained I saw frequent slowdowns, lost conn's, or loss
of service, no problems with cable. The fiber will be a lot faster for
the same cost with much lower latency, something like 2 mS if I
understand it correctly, so that will be great and probably feel more
responsive, click and stuff happens faster. The ISP includes a required
fiber-modem rental service where they can monitor or control the modem.
The typical last-hop on some of this fiber stuff, is "PON".
At the head end, one laser is split 32 ways, by a fiber splitter.
Two colors of light are used, for TX and RX. And this allows each
subscriber to only need one screw-in optical connector in the
garage. Since the two colors of light are independent, it
operates full duplex.
When they share like that, some protocol has to decide which
unit transmits next.
Using schemes like that is cheaper, for the access device at the
end of the street, but as you'd imagine, there are some fault
scenarios where 32 customers will lose service at the same time.
I had a problem with a scheme like that, at work. A shared media
network, where I hadn't put a lot of thought into the reliability
aspects. Then one day, a client unit goes nuts, and stops following
the access protocol. Which causes the whole network to go down.
That's when it occurred to me, exactly how many silicon
chips were in a single fault group. Doh! :-) I don't think anyone
in the chain of command was surprised by the result, probably
more surprised that the Devils Dice had been thrown to give
them a demo of such :-) Your fiber device is likely to be
*much* more reliable than that :-) (Crosses fingers, etc.)
They shouldn't be running "private" fiber to each subscriber,
because that would raise the installed per-channel cost. And
make the box at the end of the street, that much bigger. They
might do it that way some day, but individual fibers is
pretty expensive per channel.
And why did they pick 32 ? The optical loss goes up as the
number of channels increases. Notice how in the Specification
table for this device, for each doubling in subscribers, there's
3dB more loss. Eventually, the laser "won't make its way to the
other end" through that thing. The light is split equally from
the head end, into the fibers. There is likely to be more
loss through the splitter, than through any other component
leading to your house. The splitter is inside the box at the
end of your street.
1X2 1X4 1X8 1X16 1X32 1X64
4.1 7.2 10.4 13.4 16.4 19.9 dB insertion loss
And I have no idea what they're telling you in terms of
"available" or "guaranteed" bandwidth with such solutions.
The incoming fiber is being split 32 ways, so you get 1/32
of whatever rate that fiber runs at (worst case).
As you can tell, I'd be "full of questions" when the
installer shows up :-) I love stuff like this. Especially
when it's cheaply made and so clever.
Customers absolutely hate statistical multiplexing. I'll
never forget the "angry mob" around the Rogers booth at
the Mall, when the first cable system failed to deliver
on speed. And that's because the provider didn't have nearly
enough equipment in the core of the network, for the
number of subscribers. The mob was so angry, Rogers
closed the booth :-) So people wouldn't mill about
like angry bees. That's fixed now, and the cable network
here is every bit as competitive as any other provider.
No more angry mobs need be formed. The lesson to be learned
from this, is if you want to be a "slimeball ISP", *don't*
set up a booth at the Mall :-) Just some friendly advice.