The Ghost In The Machine <ewill@xxxxxxxxxxxxxxxxxxxxxxx> espoused:
> In comp.os.linux.advocacy, Mark Kent
> on Fri, 7 Sep 2007 22:38:34 +0100
>> The Ghost In The Machine <ewill@xxxxxxxxxxxxxxxxxxxxxxx> espoused:
>>> In comp.os.linux.advocacy, Roy Schestowitz
>>> on Fri, 07 Sep 2007 12:34:10 +0100
>>>> Feds OK Fee for Priority Web Traffic
>>>> ,----[ Quote ]
>>>> | The agency told the Federal Communications Commission, which is reviewing
>>>> | high-speed Internet practices, that it is opposed to "Net neutrality," the
>>>> | principle that all Internet sites should be equally accessible to any Web
>>>> | user.
>>> That's one way to spin it. An alternative fashion is to
>>> allow payment of a fee to expedite traffic for certain
>>> users; presumably the model will be that a website provider
>>> using a certain carrier/provider can pay extra to said
>>> carrier/provider to increase bandwidth from his website
>>> to that carrier/provider's subscribers.
>> It doesn't work, though... not without 3-mode networking.
> I'm frankly not sure what this means. Google isn't
> horribly clear on "3-mode network" either, though it did
> come up with some interesting mathematical stuff relating
> to "matrimonial ring networks". Somehow, I doubt that's
> the correct context.
Very quick overview:
1. All network traffic falls into one of three categories,
2. The three categories have key properties:
Messages and streams have urgency, files typically
do not, although most people would like them sooner rather than
later. Messages fall in between streams and files here.
2.2 Temporal relationship between symbols
In files and messages, the temporal relationship does
not matter too much, as the receiving entity reassembles
the signal anyway.
For streams, the temporal relationship is *fundamental*. The
brain cannot understand streams if they are broken up.
2.3 Symbol order (on the fly re-routing, etc.)
For streams, re-routing is not an option, as it has
unpredictable effects on the temporal aspect of the stream.
For files and messages, if packets arrive out of order, they can
be re-requested, or buffered and the file or message
3. The three networking modes are:
3.1 COCS - connection-oriented circuit-switched. This is intended for
stream connection (phone calls, videos, etc.), and can be used
to contain the other two networking modes. Examples include
PDH, SDH, WDM networks.
3.2 COPS - connection-oriented packet-switched.
This can be used for streams if the temporal relationships
and symbol order issues mentioned above can be managed. It
*cannot* be effectively used to carry COCS at other than very
low bit-rates compared to the packet rate, particularly if there
are re-transmission mechanisms built-in (like in TCP). Other
examples include ATM, FR, TCP. It is very good for messages (eg.,
telnet sessions, ssh sessions, http sessions, etc.).
For COPS to be used to carry streams, it must *not* be
transported on CLPS layers (like IP, MPLS, etc.).
3.3 CLPS - connectionless packet-switched.
This is typically simpler than COPS, thus can offer greater
throughput, but offers no routeing guarantees (unlike COCS,
above, which offers routeing guarantees). It's good for moving
files, but poor for message sessions, and is mistakenly used for
streams, ignoring the re-routeing issues. Exmaples include UDP
and IP. This *cannot* be used to transport COPS and COCS
4. Why doesn't streaming on the internet work reliably?
Fundamentally, the reason why streaming on the internet remains
unreliable, and always will be (skype is great until it doesn't work,
for example), is that trying to send streams on COPS networks is fine,
*however*, if the COPS network is transported on CLPS networks, exactly
as happens on the internet today, then the characteristics of CLPS will
eliminate any gain from adding additional functionality at the COPS
5. Can the problem be solved?
3-mode networking solves this problem, by moving files on CLPS, streams
on COCS and messages on COPS. Presently, the only type of devices
readily available to consumers which show /exactly/ this behaviour are
2.5G and 3G phones, where audio-streams are routed though the
circuit-switched network, and files and messages through the UDP and TCP
stacks instead, although the TCP stack is still subtended on the IP
layer, so it's of limited use. Further, video streams are typically
sourced on the internet, which doesn't have the infrastructure to offer
all three modes.
6. Can current generation computers support 3-mode?
Ethernet can be used to provide COPS networks which behave more or less
like COCS networks, and are therefore good enough to replace SDH layers
in telco networks, so long as it is subtended on a real COCS layer, such
as WDM. The current technology is known as PBB-TE, and has been
developed jointly by some of my work team and colleagues, and some folk in
7. What's the impact on universal access?
The debate about universal access to the internet is highly twisted
because of the mistaken belief that shovelling bandwidth at CLPS
Networks will make streaming work properly - but it won't and it can't.
The proper way for telcos to resolve this is to offer 3-mode networking,
to upgrade the internet overall so that connections can be setup for
video/audio streaming as and when required, and that the bandwidth used
is what is paid for. A poll-tax system for internet usage is not fair
on low-users at all, and encourages the consumption of large numbers of
"last-mile" links, which are hugely expensive to provide and maintain,
even if low-bandwidth usage is planned. This cost has to be borne by
| Mark Kent -- mark at ellandroad dot demon dot co dot uk |
| Cola faq: http://www.faqs.org/faqs/linux/advocacy/faq-and-primer/ |
| Cola trolls: http://colatrolls.blogspot.com/ |
| My (new) blog: http://www.thereisnomagic.org |