[Novalug] debugging low Internet throughput -- Latency on Sprint's network?

[Aaron Porter]

On Mon, Oct 26, 2009 at 2:41 PM, Paul D. Bain <paulbain at pobox.com> wrote:
>        OK, after a few minutes of fiddling, I think that I may have discovered
> something relevant. Apparently, packets traveling from Sprint's network
> to Verizon's network are experiencing considerable latency:
>
> http://www.internetpulse.net/Main.aspx?xAxis=Metric&yAxis=Origin&zAxis=Destination&nAxis=Period
>
> I may be misinterpreting this web application, however, so perhaps
> someone more knowledgeable (as to networking trouble-shooting) could
> help with interpreting the results. I have forgotten a great deal of my
> networking knowledge.

Looking at the earlier trace route, it seems that the DSL provider
runs to the Equinix Internet Exchange in Ashburn, where it peers with
peer1/serverbeach at 10gbps and then peer1 runs the packets back to
Texas over their own network. The FIOS link on the other hand runs
from Verizon (Residential) to Verizon (nee UUNet) to Sprint in New
York to tinet.net  (an Italian ISP?!?) which takes it down to Texas
and hands it off to peer1. The joys of tossing packets across the
'net.

Basically, the DSL provider has reached a (probably free) agreement
with Peer1 to pass traffic in Ashburn. Peer1 has content, the DSL
provider has eyeballs. They both win. But Verizon sees your eyeballs
as a marketable resource. They think that Peer1 should pay *them* to
have the eyeballs on their network exposed to the hosted content, as
such they probably want a ton of money from Peer1. Peer1 chose a
cheaper ISP (TI-NET) that  has some sort of traffic passing agreement
with Verizon, and that ends up being the shortest route.

A nice reminder that last-mile speeds are rarely the limiting factor.