best effort has problems

Greetings Nanogers,

I published a two month issue last weekend with the bottom line conclusion that there can be no telecom recovery as long as the industry relies solely on the best effort business model which I believe is not economically sustainable.

This has led to two articles on my June-July issue conclusions this week in the trade press. Something that has never happened to me before. :slight_smile:

The first is ISP Planet and the second Broadband Edge.

Here are the urls

http://www.isp-planet.com/perspectives/2004/cook_internet.html (monday)

http://bbedge.mblast.com/presentation/page798-878156.asp (today)

Finally my summary, table of contents and list of contributors is at http://cookreport.com/13.04.shtml

Enjoy the weekend.

I fail to see how the facts support such a conclusion. What's happening in the fiber business is free market economics 101. Sure, if people would magically start paying much more money in order to be guaranteed what they get anyway today, the people taking that money would be a lot happier. What else is new.

Still, such a "better than best effort" service has interesting technical consequences. A "better" service can't exist without a "worse" service. You can delay and drop packets a little with a lower-quality service, but not too much because TCP won't have it, leading to congestion collapse with a network full of retransmissions that never make it to the other end. So what we need is a new transport protocol or modifications to TCP that make it possible to use the available bandwidth even at high levels of congestion.

Date: Fri, 28 May 2004 15:58:06 -0400
From: Gordon Cook

I published a two month issue last weekend with the bottom
line conclusion that there can be no telecom recovery as
long as the industry relies solely on the best effort
business model which I believe is not economically
sustainable.

The PSTN doesn't offer guaranteed end-to-end transmission, and
certainly statmuxes based on expected load. Looks like similar
capacity planning.

Perhaps you refer to latency. Most people don't care as long as
HTTP and POP3 latency is "good enough" -- and server response
time is often a substantial consideration. SMTP really isn't
picky about latency or jitter.

Maybe you mean packet loss. Most everyone here can recall the
days of 30% packet loss across congested MAE FDDI fabric, but
that went away what seems like eons ago.

Eddy

The PSTN doesn't offer guaranteed end-to-end transmission, and
certainly statmuxes based on expected load. Looks like similar
capacity planning.

The PSTN does guarantee a certain service level, latency, call completion etc.

Perhaps you refer to latency. Most people don't care as long as
HTTP and POP3 latency is "good enough" -- and server response
time is often a substantial consideration. SMTP really isn't
picky about latency or jitter.

Latency & Jitter are very important when dealing with sound & video. Or anything realtime for that matter. The Internet isn't just HTTP, NNTP, SMTP any more.

Maybe you mean packet loss. Most everyone here can recall the
days of 30% packet loss across congested MAE FDDI fabric, but
that went away what seems like eons ago.

I remember quite a bit of packet loss when the last series of worms hit

Date: Sat, 29 May 2004 14:26:01 -0400
From: Matthew Crocker

The PSTN does guarantee a certain service level, latency,
call completion etc.

As do many Internet providers. (s/call completion/packet loss/)

Latency & Jitter are very important when dealing with sound &
video. Or anything realtime for that matter. The Internet
isn't just HTTP, NNTP, SMTP any more.

Nitpicking: Latency isn't that important with unidirectional
communication. However, VoIP users seem reasonably happy with
current latency and jitter -- and the Internet still is _largely_
xxTP, anyway... particularly if one ignores peer-to-peer file-
swapping programs.

I remember quite a bit of packet loss when the last series of
worms hit

As do I, but I consider that an exception, not part of standard
operation. Admittedly, that may well be an error on my part,
considering the increasing popularity of worms and viruses. The
PSTN doesn't have botnets of "pwned" phones making prank calls.
(Further note that MAE FDDI congestion was more frequent than the
current malware field days.)

However, I see this as a problem of securing machines, not one of
best effort delivery. Choke trunks are used to prevent radio
call-ins from overloading the PSTN.

Perhaps throttling bandwidth using a slow-moving exponential
decay, over a long averaging period, is a good idea. One could
allow short bursts of line-rate traffic.

End-user duty cycles are low. This is what facilitates current
levels of statmuxing, and why packet loss skyrockets when many
systems try to operate at line rate for extended durations.

Eddy

Latency is fine for VOIP as long as you dont interact with the PSTN
network, if you want to interact with PSTN then you need echo
cancellation if you have high latency on the IP part.

Most VOIP applications can handle 40ms jitter, so that's normally no
problem unless your local access is full. Packet loss is normally no
problem for VOIP if you use a proper (non-telco developed) codec.

VOIP is actually better off with high packet loss and low jitter than the
other way around (throwing off the old truth that core equipment should
have lots of buffers).