To address this, Nichols and Jacobson
have resumed work on a robust, adaptive AQM.
Be Part of the solution
The situation may worsen before it improves, and immediate action is necessary. Potential solutions must be subjected to rigorous testing and analysis
before being widely deployed; otherwise, existing problems can be made
worse. Unfortunately, today there is a
distinct lack of funding for the kinds
of performance monitoring, tuning,
and improvement that characterized
the early Internet.
The first step is to make the problem
apparent. Consumer tests are important (for example, Speedtest.net, Sam-Knows, M-Labs (
http://www.measure-mentlab.net/), Netalyzr), but better
tests that point to the correct offender,
usable by everyone, are badly needed. Consumer tests often perpetuate
the mythology that more bandwidth
means higher “speed,” and better marketing metrics are essential. Stuart
Cheshire’s famous “It’s the Latency,
Stupid” rant should be taken to heart. 3
An open source project, CeroWrt,
is under way at bufferbloat.net using
OpenWrt to explore potential solutions, including AQM. Please help.
A wide range of testing is needed for
confidence in any algorithm. Since our
operating systems are commodities
and are used in today’s home routers,
home-router bufferbloat is a direct result of host bufferbloat. Solve one, and
you solve the other.
Unfortunately, since bufferbloat
misleads TCP’s congestion-avoidance
algorithm with respect to the effective
pipesize, modern networks without
effective AQM may again be vulnerable to congestion collapse from saturated edge buffers creating packet
delays measured in seconds. Congestion collapse has been reported in a
large-scale network, requiring complete shutdown and careful restart of
the entire network to regain (
We are flying on an Internet airplane in which we are constantly swapping the wings, the engines, and the
fuselage, with most of the cockpit instruments removed but only a few new
instruments reinstalled. It crashed before; will it crash again?
buffers is at
the path bottleneck,
but the critical
be different for
the reverse path,
and with dynamic
the same path.
We would like to thank Dave Clark,
Dave Reed, Vint Cerf, Van Jacobson,
Vern Paxson, Nick Weaver, Scott Bradner, Rich Woundy, Greg Chesson, Dave
Täht, and a cast of hundreds.
George V. Neville-Neil
network Virtualization: Breaking
the Performance Barrier
You Don’t Know Jack about
Kevin Fall, Steve McCanne
1. appenzeller, g., keslassy, I. and Mckeown, n. sizing
router buffers. In Proceedings of ACM SIGCOMM
(Portland, aug. 2004).
2. braden, r. et al. recommendations on queue
management and congestion avoidance in the
Internet, rfC2309 (april 1998).
3. Cheshire, s. It’s the latency, stupid (1996); http://
4. dischinger, M. et al. Characterizing residential broadband
networks. In Proceedings of Internet Measurement
Conference (san diego, Ca, oct. 24–27, 2007).
5. floyd, s. and jacobson, v. random early detection
gateways for congestion avoidance. Ieee/aCM
transactions on networking (aug. 1993).
6. jacobson, v. notes on using red for queue
management and congestion avoidance. talk at north
american network operators’ group (1998) 13; ftp://
7. jacobson, v. notes on using red for Queue
Management and Congestion avoidance; ftp://ftp.
ee.lbl.gov/talks/vj-nanog-red.pdf, see also http://www.
8. jacobson, v. and karels, M. Congestion avoidance and
control. In Proceedings of SIGCOMM ’ 88, (aug. 1988).
9. kreibich, C. et al. netalyzr: Illuminating the edge
network. In Proceedings of the Internet Measurement
Conference (Melbourne, australia, nov. 1–3, 2010).
10. nagle, j. on packet switches with infinite storage.
In Proceedings of Network Working Group RFC 9790
(dec. 1985); www.ietf.org/rfc/rfc970.txt.
11. reed, d. P. Congestion collapse definition (2009);
12. reed, d. P. what’s wrong with this picture (2009);
13. rhee, I. and Xu, l. CubIC: a new tCP-friendly high-speed tCP variant. ACM SIGOPS 42, 5 (2008).
14. villamizar, C. and song, C. high-performance tCP in
ansnet. Computer Communications Review 24, 5
Jim Gettys is at alcatel-lucent bell labs, usa, where he
works on bufferbloat, as a properly working low-latency
Internet is required for immersive teleconferencing. he
was vice president of software at the one laptop per Child
project, editor of the httP/1.1 standard, and is one of the
original designers of the X window system.
Kathleen nichols is the founder and Cto of Pollere Inc.,
a consulting company working in both government and
commercial networking. she has 30 years of experience
in networking, including a number of silicon valley
companies and as a cofounder of Packet design.
© 2012 aCM 0001-0782/12/01 $10.00