[e2e] Re: crippled Internet

Jon Crowcroft J.Crowcroft at cs.ucl.ac.uk
Wed Apr 25 08:19:53 PDT 2001

walled gardens

i can measure the edge2edge performance on lots of networks and find
the typical tier 1 type statement but the provisioning of many
inter-domain points is not exactly keeping up - there's some mileage
in following metcalfes law for connectivity (value of the internet goes up
as square of number of things interconnected) but there's also price
diffeentiation (value of loss free path between preferred
business customers of , whether intentional or not, is higher) so you
get an end2end picture that encompasses lots of market mismatches,
whether deliberate, short term, accidental, long term or whatever

some of these things are also measures of the slowness of "convergence" -
despite imminent danger of deployment, we stil havnt seen enough
fallout in the price of trans-pond capacity quite yet 

and then again a lot of access providers are genuinely struggling to
deploy fast enough

finally, there;s always server underpovisioning - a lot of the time,
if you look at the response time from popular (xxx:-) servers
it is an order of magnitude slower than the time to get the DNS lookup
and subsequent TCP connection setup exchange done - the NY times
doesnt care too much if its the http resonse (or dns cache miss), 
or the TCP good put, or the
IP queueing and packet loss that caused the problem....i guess we need
a pie chart:-)

In message < at mira-sjcm-2.cisco.com>, Fred Bake
r typed:

 >>You folks disagree with my analysis; maybe you have a better explanation 
 >>than I do. What is your explanation? Here's the problem.
 >>Upper crust ISPs generally throw enough bandwidth into their networks and 
 >>between their networks that delay and loss are nominal. When I talk with 
 >>you guys about problems in the net, you say "show me the link, because I 
 >>manage my links at some huge over-capacity".
 >>When users measure Internet behavior, they say that delay is highly 
 >>variable and loss is excessive. We get all manner of stuff like the NY 
 >>Times article the other day which said that if the Internet didn't develop 
 >>something resembling QoS, it would be bypassed. The Internet Weather Report 
 >>folks have quite a going business helping them measure this, and as you 
 >>know, they really *do* like to bash the ISP.
 >>I don't see any evidence that either is stupid or incorrect in their 
 >>reports. They seem to live on different planets, though. Tell me where the 
 >>problem lies if it is not on the interconnect between them.



More information about the end2end-interest mailing list