Q on FCC speed tests

[David Farber <dave () farber net>]

Begin forwarded message:

From: Karl Auerbach <karl () cavebear com>
Date: March 12, 2010 8:15:46 PM EST
To: dave () farber net
Cc: ip <ip () v2 listbox com>
Subject: Q on FCC speed tests
Reply-To: karl () cavebear com


Some things that I did not understand about the FCC "broadband" test are these.  Maybe I didn't read deep enough?

What part of the net are they measuring?  Is it the characteristic of the first IPv4/IPv6 hop?  Or source to the second 
IP router?  Or source to the third IP router?  Hop N?  Or are they going to some canonical target?

Are they doing it the simple, dumb way or are they doing a detailed hop-by-hop analysis over a period of time using 
techniques pioneered by Van Jacobson in "pathchar" and refined in subsequent tools like "pchar"?  (From what I saw when 
I ran it it appeared that they were doing it using the simple, dumb way.)

What MTU size?  (Some access paths don't support full 1500 byte MTU, which can be a performance disaster for users 
whose software assumes 1500 or does not do Path MTU discovery.)  And what about access nets, if any, that might support 
jumbograms?

What QoS settings?

What about path asymmetry or load-balancing across parallel paths?

Do they measure things that can really cause protocol stacks to have hysterics, things like reordering?

And do they measure the degree of buffering in the path?  Some gear, particularly consumer gear, often contains a lot 
of buffering that can affect measurements.

And thinking of buffering, do the tests evaluate the kind of discard policies on the path?  Is the discard application 
protocol or packet size agnostic?  Is it tail drop, head drop, random drop?  Does the drop policy try to apply policies 
such preferring to drop TCP packets over UDP packets, and preferring those TCP packets not containing ACK, SYN, or FIN 
flags?

How a path reacts to flow dynamics can be very important to a user's perceptions.  VoIP is a nice steady, metronomed 
sequence of smallish packets while some forms of high-grade video consist of burst-trains of closely spaced large 
packets with the bursts occurring several times a second.

Path evaluation and comparison isn't something that can be done without considering the use to which the path will be 
put.  For example, in my "Fast Path Characterization Protocol" idea - 
http://www.cavebear.com/archive/fpcp/fpcp-sept-19-2000.html - I carry a hypothetical IP header and description of the 
flow dynamics in the path measurement.

                --karl--





-------------------------------------------
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com