Communications - July 2012

user
pkts
Scheduler
Ethernet
( 100 Mbps–1Gbps)

CM buffer

Cable Modem

upstream (2Mbps-?)

When uplink is heavily utilized
the router buffers will fill
Rate mismatch will
cause CM buffer to fill
(bufferbloat) and may
cause priority inversion

a CeroWrt-enabled edge router to make bufferbloat disappear. Unfortunately, large buffers are not always located where they can be managed but can be ubiquitous and hidden. 6, 7 Examples include consumer-edge routers connected to cable modems and wireless access points with ring buffers. Many users access the Internet through a cable modem with varying upstream link speeds: 2 Mbps is a typical value. The home network or home computer connects to the cable modem with an Ethernet cable in the range of 100Mbps–1Gbps ( Figure 10). The modem’s buffers are at the fast-to-slow transition, and that is where queues will build up: inside a sealed device outside of user control. Any differentiated services (DiffServ) queuing and buffer management at the router can be defeated by a full queue in the cable modem. Three approaches have been suggested: limit the Ethernet link to the upstream rate; put buffer management and DiffServ queues in the cable modem with a configuration interface for the DiffServ queues; and implement Ethernet flow control between the modem and the upstream router.

Option 1 has the advantage that it
can be implemented by a user with-
out cable-modem changes and the
disadvantage that it must rate limit to
the expected rate of the upstream. If
the rate drops, then a queue will still
build up in the cable modem, and any
additional short-term bandwidth can-
not be utilized. Option 2 puts buffer
management right at the bottleneck
link but requires the vendor to make
(possibly significant) changes to the
modem architecture and permit con-
figuration. Option 3 also requires
vendors to make changes but uses
Ethernet flow control to permit only
the number of packets in the modem
buffer needed for good transmission
utilization while pushing the queue
into the router where it can be man-
aged and where new algorithms can
be more readily deployed. Options 2
or 3 are preferable but require a cable-
modem vendor and/or a cable data
network service provider to make this
part of the modem requirements.

next steps The open source project CeroWrt3 is using Open Wrt to explore solutions to bufferbloat. A CoDel implementation is in the works, after which real-world data can be studied. We plan to make our ns- 2 simulation code available, as well as some further results. 19

References

1. braden, r. et al. recommendations on queue management and congestion avoidance in the internet. rFC 2309 (1998).

2. bufferbloat Project; http://www.bufferbloat.net.

3. Cero Wrt Project; http://www.bufferbloat.net/projects/ cerowrt.

4. dischinger, M. et. al. Characterizing residential broadband networks. in Proceedings of the Internet Measurement Conference, san diego, Ca, 2007.

5. Floyd, s. and Jacobson, v. random early detection gateways for congestion avoidance. IEEE/ACM Transactions on Networking, (1993).

6. gettys, J. bufferbloat: dark buffers in the internet. backspace Column. IEEE Internet Computing 15, 3 (2011), 95-96.

7. gettys, J. and nichols, K. 2011. bufferbloat: dark buffers in the internet. Commun. ACM 9, 11 (sept. 2011) 57–65.

8. Jacobson, v. Congestion avoidance and control. Proceedings of SIGCOMM ’ 88, (stanford, Ca, 1998).

9. Jacobson, v. reported in Minutes of the Performance Working group. Proceedings of the Cocoa Beach Internet Engineering Task Force. Corporation for national research initiatives, reston, va, 1989.

10. Jacobson, v. notes on using red for queue management and congestion avoidance. talk presented at north american network operators’ group (1998); ftp://ftp.ee.lbl.gov/talks/vj-nanog-red.pdf.

11. Jacobson, v. a rant on queues. a talk presented at Mit lincoln labs, lexington, Ma, 2006; http://www. pollere.net/Pdfdocs/QrantJul06.pdf.

12. Jacobson, v., nichols, K. and Poduri, K. red in a different light, 1999; http://www.cnaf.infn.it/~ferrari/ papers/ispn/red_light_ 9_ 30.pdf.

13. Kreibich, C. et. al. netalyzr: illuminating the edge network. in Proceedings of the Internet Measurement Conference, (Melbourne, australia, 2010).

14. li, t. and leith, d. adaptive buffer sizing for tCP flows in 802.11e Wlans. in Proceedings of the 2008 Communications and Networking in China.

15. Mankin, a. random drop congestion control. in Proceedings of SIGCOMM ’ 90.

16. Mathis, M., semke, J. and Mahdavi, J. the macroscopic behavior of the tCP congestion avoidance algorithm. ACM SIGCOMM Computer Communication Review 27, 3 (1997).

17. nagle, J. Congestion control in iP/tCP internet works. rFC 896 (1984); http://www.ietf.org/rfc/rfc896.txt,.

18. network simulator - ns- 2; http://nsnam.isi.edu/nsnam/ index.php/user_information.

19. http://www.pollere.net/Codel.html.

20. vu-brugier, g., et. al. 2007. a critique of recently proposed buffer-sizing strategies. ACM SIGCOMM Computer Communication Review 37( 1).

21. Weigle, M. C. 2002. Web traffic generation in ns- 2 with PackMime-httP; http://www.cs.odu.edu/~mweigle/ research/packmime.

22. Feng, W. et. al. 2002. the blue active Queue Management algorithm. IEEE/ACM Transactions on Networking 10, 4 (2002), 513–528.

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.

Van Jacobson is a research Fellow at ParC where he leads its content-centric networking research program. he has worked at lawrence berkeley national laboratory, Cisco systems, and was a cofounder of Packet design.