|
| |
Distributed congestion control
for heterogeneous networks
Sponsor: NSF
Abstract
In light of TCP’s scalability
issues in high bandwidth-delay networks, explicit-feedback congestion control
has gained renewed interest in the last several years. Sometimes referred to as
Active Queue Management (AQM) congestion control, these algorithms rely on
routers to provide the various types of congestion feedback including changes to
the congestion window, packet loss, single-bit congestion indication, queuing
delay, and link prices. This project aims to explore the various properties of
the prior AQM methods, propose new controllers that improve the performance of
existing approaches in highly heterogeneous network conditions, and evaluate
their
performance in a network of software and hardware routers.
Journal
Papers
| |
Y. Zhang and D. Loguinov, "ABS: Adaptive
Buffer Sizing for Heterogeneous Networks," Elsevier Computer
Networks, vol. 54, no. 14, October 2010.
|
PDF |
|
|
Y. Zhang, Y. Xiong, S. Liu, and D.
Loguinov, "Queuing Dynamics and Single-Link Stability of Delay-Based
Window Congestion Control," Elsevier Computer Networks, vol. 54,
no. 10, July 2010.
|
PDF |
|
|
Y. Zhang, S. Jain, and D. Loguinov,
"Towards Experimental Evaluation of Explicit Congestion Control," Elsevier Computer Networks,
vol. 53, no. 7, May 2009.
|
PDF |
|
|
Y. Zhang and D. Loguinov, "On
Delay-Independent Diagonal Stability of Max-Min Congestion Control,"
IEEE Transactions on
Automatic Control, vol. 54, no.
5, May 2009.
|
PDF |
|
|
Y. Zhang and D. Loguinov,
"Local and Global Stability of Delayed Congestion Control Systems,"
IEEE Transactions on Automatic Control, vol. 53, no. 10, October 2008.
|
PDF |
| |
Y. Zhang, D. Leonard, and D. Loguinov,
"JetMax: Scalable Max-Min Congestion Control for High-Speed
Heterogeneous Networks," Elsevier Computer Networks, vol. 52, no.
6, April 2008.
|
PDF |
| |
Y. Zhang, S.-R. Kang, and D.
Loguinov, "Delayed Stability and Performance of Distributed Congestion
Control," IEEE/ACM Transactions on Networking, vol.
15, no. 4, August 2007.
|
PDF |
Conference
Papers
|
|
Y. Zhang and D. Loguinov, "ABS: Adaptive
Buffer Sizing for Heterogeneous Networks," IEEE IWQoS, June 2008.
|
PDF, PPT |
|
|
S. Jain, Y. Zhang, and D. Loguinov,
"Towards Experimental Evaluation of Explicit Congestion Control," IEEE
IWQoS, June 2008.
|
PDF,
PPT |
| |
S. Bhandarkar, A.L.N. Reddy, Y. Zhang, and
D. Loguinov, "Emulating AQM from End Hosts," ACM SIGCOMM, August 2007.
|
PDF |
| |
S. Jain and D. Loguinov, "PIQI-RCP:
Design and Analysis of Rate-Based Explicit Congestion Control,"
IEEE IWQoS, June 2007.
|
PDF,
PPT |
| |
Y. Zhang and D. Loguinov, "On
Delay-Independent Diagonal Stability of Max-Min Congestion Control,"
IEEE CDC, December 2006.
|
PDF,
PPT |
| |
Y. Zhang, D. Leonard, and D. Loguinov, "JetMax:
Scalable Max-Min Congestion Control for High-Speed Heterogeneous
Networks," IEEE INFOCOM, April 2006.
|
PDF,
PPT |
| |
Y.
Zhang and D. Loguinov, "Local and Global Stability of
Symmetric Heterogeneously-Delayed Control Systems," IEEE
CDC, December 2004.
|
PDF,
PPT |
| |
Y.
Zhang, S-R. Kang, and D. Loguinov, "Delayed Stability and
Performance of Distributed Congestion Control," ACM SIGCOMM,
August 2004.
|
PDF,
PPT |
Detailed Project Description
MKC
Max-min Kelly Control (MKC) is
an active queue management (AQM) congestion control for high-speed networks.
MKC collects feedback from the most-congested router along each flow's path and achieves many desirable
properties of congestion control including:
-
Quick convergence to link utilization. MKC takes the same number of RTT steps to fully utilize a 10 mb/s, 10 gb/s, or googol bps link. For the parameters shown in the
MKC
paper, 99% utilization is reached in 6-7 RTTs for any link capacity.
-
RTT-independent fairness.
All flows sharing the same bottleneck receive an equal share of the
bandwidth, regardless of their round-trip delay or path length.
-
Asymptotic stability under consistent bottleneck assignment and all
types of delay. Flow rates converge to the equilibrium without
oscillation under both heterogeneous and time-varying feedback delays.
-
Insensitivity to reverse-path
congestion. MKC is not sensitive to the delay dynamics
of the reverse path.
-
Low overhead. MKC requires one
addition per arriving packet inside routers and utilizes a fixed control interval length
for generating feedback. The MKC packet header is only 20 bytes.
JetMax
JetMax is an improvement of MKC that eliminates
its steady-state
packet loss, allows tunable link utilization, and improves the convergence rate
to fairness. It inherits the above properties of MKC and additionally exhibits
the following features:
-
Zero packet loss. Loss-free
operation is ensured both in the transient and stationary state.
-
Tunable link utilization. Each router can be independently configured such that the flows bottlenecked
at that router automatically achieve arbitrary link utilization.
-
Quick convergence to max-min fairness. Fairness is reached in a fixed number of RTT steps regardless
of link capacity or number of flows in the network.
-
Low overhead. JetMax requires three additions per arriving packet inside
routers and 28 bytes in the header. It also allows simple integration into
Linux and Windows 2003 kernels.
Code
Ns2 simulation code (MKC,
JetMax)
Ns2 scripts for the papers (MKC,
JetMax)
Emulab Linux drivers (JetMax, RCP, XCP, MaxNet)
JetMax Windows driver (experimental)
|