110 likes | 250 Views
Using Jitter As A Predictor Of Congestion. Logan Kennelly. Motivations. Real-Time Streams Codecs avoid congestion effects Don’t respond to congestion Non-elastic applications Could reduce payload Could reduce overhead. Goals. Detect congestion… “In-band” Utilize data from the media
E N D
Using Jitter As A Predictor Of Congestion Logan Kennelly
Motivations • Real-Time Streams • Codecs avoid congestion effects • Don’t respond to congestion • Non-elastic applications • Could reduce payload • Could reduce overhead
Goals • Detect congestion… • “In-band” • Utilize data from the media • Before it occurs • Predict congestion before losses
Existing Methodologies • TCP-Like Congestion Control[1] • Responds to losses, assumes elastic application • PathLoad[2]/PathMon[3] • Bandwidth estimation by packet trains • Explicit Congestion Notification[4] • Probably best, but not end-to-end
On The Use Of Jitter • delaytotal = delaytransmission + delayprocessing + delayqueuing • Increased congestion -> increased queuing delay • For fixed-size packets, transmission and processing delays should be fixed • Thus, a change in total delay could be interpreted as congestion
Assumptions • Bandwidth available, only short-term congestion • Fixed route • Varying path characteristics can influence jitter • Fixed-size stream packets
Real-time Transport Protocol[5] • Timestamps every sent packet • Easily calculate jitter: (tri - tri - 1) - (tsi - tsi - 1) • Includes provisions for stream reports • Detect congestion at receiver • Note: Current work doesn’t report back
Methodology • End hosts • Single sender/receiver pair • Data rate is 9.3 kB/s • Comparable to a real-time audio stream including overhead • Sender sends at constant 30 ms intervals • Receiver records the jitter • I process the data after simulation
Methodology • ns-2 Network • Five routers • Competing Pareto-distributed traffic • 500 ms burst intervals • One bottleneck
Preliminary Results • Jitter appears to be merely an acceptable predictor • Measurements for this network appear to be mostly random • Haven’t quantitatively tested for randomness • Brief increase in jitter just before packet loss event • Also increases on non-loss events
References [1] M. Allman, V. Paxson, and W. Stevens. TCP congestion control, April 1999. Status: PROPOSED STANDARD. [2] M. Jain and C. Dovrolis. Pathload: A measurement tool for end-to-end available bandwidth. In 3rd Passive and Active Measurements Workshop, March 2002. [3] D. Kiwior, J. Kingston, and A. Spratt. Pathmon, a methodology for determining available bandwidth over an unknown network. In Advances in Wired and Wireless Communication, pages 27–30, April 2004. [4] K. Ramakrishnan, S. Floyd, and D. Black. RFC 3168: The addition of explicit congestion notification (ECN) to IP, September 2001. Status: PROPOSED STANDARD. [5] H. Schulzrinne, S. Casner, R. Frederick, and V. Jacobson. RTP: A transport protocol for real-time applications, July 2003. Status: STANDARD.