links on the blink (fwd)

10% packet loss is quite within the normal range of parameters for a
packet switching network such as the Internet.

Well, I don't think I will need to be reminded not to buy Internet service
from your organization.

Here is a bit of context.

At home I am a beta site for the PSI Internet service via cable TV. I
generally use a computer at Harvard University to read my mail. Although
the Harvard network is about 1.5 miles away from my house the sparse
distribution of connections between Internet service providers and the
current routing policies mean that packets from my house to Harvard take 20
hops and go through San Francisco and packets from Harvard to my house,
(also 20 hops) go through Washington DC.

Here are the results of some reliability tests I just ran.

    209 packets transmitted, 209 packets received, 0% packet loss
    round-trip min/avg/max = 99.996/268.978/983.334 ms

from Harvard to Washington DC via BBN Planet and MCI

    204 packets transmitted, 204 packets received, 0% packet loss
    round-trip (ms) min/avg/max = 16/46/249

BBN Planet, MCI, MAE-east and PSI.

    505 packets transmitted, 497 packets received, 1% packet loss
    round-trip min/avg/max = 116.702/192.69/3000 ms

(Note this is being tested at a time when Harvard's service provider (BBN
planet) is busily working on one of the links in its backbone because they
do not think the quality is high enough.)

I consider it a problem when the loss exceeds 1% through this long path -
as do the people who run the networks that my traffic passes through. The
normal loss through this path is less than 1% and, much of the time it is 0.

But nothing is broken. There is no inter-ISP level.

It is so much easier to just say it is the other guy's problem.
Hans-Werner suggests that most phone companies do not take this attitude,
lucky for some of us, not all Internet "service" providers do.

Scott

(just in case you might think I'm off on some other world, there are other
paths I could have used which would have shown far higher loss, those paths
would have gone through service providers who apparently feel the way that
the person quoted here does or who are unable to make the investment in
their infrastructure warranted by the level of traffic their customers
would like to exchange.)

> 10% packet loss is quite within the normal range of parameters for a
> packet switching network such as the Internet.

Well, I don't think I will need to be reminded not to buy Internet service
from your organization.

Why not? We don't run the Internet, we just connect people to it. And our
10Mbps fibre ATM link rarely gets to 25% utilization.

I consider it a problem when the loss exceeds 1% through this long path -
as do the people who run the networks that my traffic passes through. The
normal loss through this path is less than 1% and, much of the time it is 0.

momentary samples such as yours are meaningless. One moment you can have
0% loss, the next moment 10%. I will agree that sustained high loss
levels are not acceptable, but isolated ping measurements do not measure
that.

> But nothing is broken. There is no inter-ISP level.

It is so much easier to just say it is the other guy's problem.
Hans-Werner suggests that most phone companies do not take this attitude,

I disagree. If you get busy signals or recorded messages from a telco
other than the one who provides your local phone service, the most they
will do is to promise you that they will look into it and/or inform the
company whose service appears to be the problem. This is all any ISP can
do when a customer complains that ISP X is not reachable or that ISP X's
site appears to be overloaded. I am talking about ISP's here, not NSP's.

Michael Dillon Voice: +1-604-546-8022
Memra Software Inc. Fax: +1-604-542-4130
http://www.memra.com E-mail: michael@memra.com