🚀 go-pugleaf

[POSTED TO comp.dcom.modems.cable; PLEASE REPLY THERE]

In <3A78327F.BDD53DB3@mynetwatchman.com>, Lawrence Baldwin
<baldwinlSPAMMENOT@mynetwatchman.com> wrote:

>I thought that even the very first (pre DOCSIS) cable modem system used
>CSMA/CD.  I also though that even DOCSIS reverts to a CSMA/CD-type
>access method if you diable the slotting processes.

DOCSIS provides that upstream utilization is controlled by the CMTS
(headend) -- the cable modem (CM) competes (contends) for capacity, but
that capacity is granted (allocated) by the CMTS.  For details, see:

   Data-Over-Cable Service Interface Specifications
   Radio Frequency Interface Specification
   SP-RFI-I05-991105
   <http://www.cablemodem.com/SP_RFI_I05_991105.pdf>
   6.4 UPSTREAM BANDWIDTH ALLOCATION

>No, what I'm saying is that with CSMA/CD you have ONE shared upstream
>channel, thus saturation would
>be easier to occur, so I'd be even more concerned.

Saturation of the uplink is the same issue regardless of the number of
channels.  What matters is aggregate demand versus aggregate available
capacity.

>What I think John is saying is that DOCSIS cable allocates upstream
>slots to each user.  Slotting
>should provide some minimal amount of bandwidth to each user, however if
>the total
>slot allocation is so over-subscribed that a few (dozen?) users are
>using their total allocation
>can cause upstream saturating.
>
>The key comes down to just how oversubscribed is the upstream?
>
>If you've got 500 users with 256,000bps contending for 2,000,000 that's
>> 50:1...which is
>2x-3x what ADSL typically is.  These are just numbers I'm pulling from
>the air, I defer to someone
>that knows cable infra to post real numbers.

Cisco's recommendation for the uBR7200 series (CMTS) is 1000 cable
modems per downstream port (as compared to a theoretical DOCSIS maximum
of 8175).

The downstream channel can be 27 Mbps (QAM 64) or 36 Mbps (QAM 256).  My
understanding is that 27 Mbps (QAM 64) is the most common configuration
in practice.

If we assume the service objective to be 1.5 Mbps downstream, then the
oversubscription rate is about 55:1 (1000 x 1.5 / 27).  My understanding
is that this rate is often exceeded in practice as penetration increases
-- oversubscription rates of 100:1 are commonly discussed.

The upstream channel can be anywhere from 160 Kbps (200 kHz QPSK) to 10
Mbps (3.2 MHz QAM 16).  My understanding is that 800 Khz QPSK is most
common in practice (due to signal quality and error rate issues), giving
an upstream channel capacity of 640 Kbps.  With 4 upstream channels per
downstream channel, the aggregate upstream capacity would be about 2.5
Mbps, as compared to downstream capacity of 27 Mbps.  Note that if the
upstream saturates, the downstream rate will drop to about the same
speed, a dramatic slowdown of about 90% (2.5 Mbps as compared to 27
Mbps)!

Even with cable modems capped to 128 Kbps upstream, that 2.5 Mbps
upstream capacity can handle only 20 simultaneously active modems before
saturation.  That's not a problem if cable modem usage is typically (1) 
infrequent, (2) downstream [e.g., web surfing], and (3) interactive
[e.g., fetch-use].  The system can break down if those conditions are
not met.

This makes it easier to see why many cable MSO's condemn *continuous*
use of upstream (e.g., running a popular public server) as "abuse" --
each such modem consumes capacity normally allocated for 1000 / 20 = 50
subscribers.  Worse, there's a threshold effect:  If the upstream is
running at (say) 80% of capacity with typical subscribers, it takes only
4 (out of 1000+) heavy upstream users at 128 Kbps to drive the upstream
into saturation, thereby slowing downstream to a crawl for *all*
subscribers on that segment.  (Note that!)

Even with double the upstream capacity (e.g., 4 channels of 1.6 MHz QPSK
= about 5 Mbps), the upstream saturates at only 40 simultaneously active
modems.  Although downstream slowdown due to upstream saturation is less
severe, it's still about 80% (5 Mbps as compared to 27 Mbps)!

So-called "fairness" doesn't enter into this because it doesn't come
into play until the upstream saturates, but then it's too late.  Given
the bursty nature of typical upstream traffic (fetching web pages from
time to time), "fairness" won't throttle heavy upstream users in a way
that avoids saturation.  "Fairness" does prevent any one user from
hogging a disproportionate share of the uplink (at any given instant in
time), but it doesn't prevent the uplink from saturating.

The bottom line is that the weakness of cable (as compared to DSL) is
that limited upstream capacity is shared among all subscribers on a
segment, making them *all* vulnerable to saturation of the upstream by a
relatively small number of upstream users.  Dealing with the problem by
deploying more upstream capacity is relatively expensive; e.g., even
with an upstream cap of only 128 Kbps, upstream capacity of 2.5 Mbps
serves only 20 heavy upstream users as compared to 1000 typical
subscribers (2%).  It's therefore not surprising that so many cable
MSO's focus instead on eliminating heavy upstream use.  (That is of
course their right -- this is really about honesty, not fairness.)

-- 
Best regards,
John Navas     <http://navasgrp.home.att.net/>
     CABLE MODEM/DSL GUIDE:  <http://Cable-DSL.home.att.net/>