[funsec] "Network Neutrality" or "Open Internet"

Fergie fergdawg at netzero.net
Sun Apr 16 09:14:12 CDT 2006

One more thing: A good primer & prerequisite for understanding
QoS in packet communications is reading Len Kleinrock's works
on queuing.

It's all about queuing. ;-)

- ferg

-- "Fergie" <fergdawg at netzero.net> wrote:

-- "Dude VanWinkle" <dudevanwinkle at gmail.com> wrote:

>> In general terms, remember that electrical or optical signalling
>> is _clocked_ into 'the network'.

>err, what?

Think of it this way: Your circuit 'size' is just timing, re:
1.536Mb/s. See? It's just electrical clocking. It's a matter of
speed, not size.

It does not have any other function.

Routers, on the other hand, recieve and transmit data at their
(electrical or optical) associated _speeds_ on various interfaces,
but at the electrical level (go back and re-acquait yourself with
the OSI 7-layer model, for example), but it has absolutely no
effect on packet communications aside from clocking bits onto a

The router still has to look at the bits, determined header and
encapsulation information, match them to destinations (remember that
all forwarding in IP is done by destination), and put them back out
on the appropriate interface, and then you are back at electrical
(or optical) clocking (speed) again.

Where this gets interesting is with overcommitment, as Valdis pointed
out. You can't put ten pounds of shite in a five pound bag, so when
there are times of day when the aggregate amount of traffic is graeter
than the router is able to put on the wire at any given time, the
queues start to fill up, and packets will be dropped.

Now, dependent on the amount of packet drops is how the transport
protocols behave during loss -- TCP has slow-start, UDP (by design,
unfortunately) keeps spitting traffic into the network and relies
on the application layer to detect and adjust to packet loss.

When the amount of packet loss becomes arbitrarily large, then it
becomes an exercise for the reader to determine how to 'manage'
the packet loss -- normally by doing some creative packet drop
based on what traffic you think is 'important' or who pays you
more not to drop their traffic.

Of course, there is also fanciful non-FIFO queuing schemes, but
I won't go into that now. :-)

- ferg

p.s. SLA's are _almost_ always written for _availability_ -- not
capacity or throughput -- THAT is handled by the circuit _speed_,
which is always constant. ;-)

Confused yet? ;-)

"Fergie", a.k.a. Paul Ferguson
 Engineering Architecture for the Internet
 fergdawg at netzero.net or fergdawg at sbcglobal.net
 ferg's tech blog: http://fergdawg.blogspot.com/

More information about the funsec mailing list