Download presentation
Presentation is loading. Please wait.
1
Computer Science 1 Characterizing Link Properties Using “Loss-pairs” Jun Liu (joint work with Prof. Mark Crovella)
2
Computer Science 2 Problem Definition Discovering network link properties is crucial to network-adapting applications and services: adapting congestion control behaviors of traffic sources sharing the same bottleneck link. predicting the delay and jitter for QoS and real-time applications. “loss pairs”: new method for discovering link properties such as buffer size, active queue management policies in use
3
Computer Science 3 The Main Idea A “loss-pair” is a pair of packets traveling the same network path sent close together in time, such that the first packet is dropped (but not the second) The main idea is that the second packet experiences network conditions similar to that seen by the first packet In this way, correlated information about packet dropping behavior and queue occupancy can be obtained at an end system
4
Computer Science 4 Evaluation Scenario We intend to characterize the link property between node B and C by the measurements taken at the observer.
5
Computer Science 5 Experimental Results To show the validity of our method, we measured the rtt of the second packet in the loss-pair for increasing buffer size of DropTail bottleneck queue.
6
Computer Science 6 Experimental Results (2) Loss-pairs can also be used to characterize the dropping behavior of the bottleneck queue, we did experiments on DropTail, RED and BLUE queues to show their dropping functions. REDBLUE
7
Computer Science 7 Conclusion Loss-pair is a promising measurement tool that can be used to characterize the dropping behavior of a network path furthermore, it can also be used to estimate parameters associated with the bottleneck queue. Future work: To do real world measurement to further test the validity of this technique, and think about its extension to estimate more useful network-internal information.
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.