1. What do packet dispersion techniques measure? Internet Systems and Technologies - Monitoring

2. Contents  Introduction  Packet pair dispersion  Cross traffic interference  Packet train dispersion  The size of probing packets  The “pathrate” methodology  Heuristic evaluation  Conclusion

3. Introduction  Why we need bandwidth monitoring?  Metrics  Capacity (no cross traffic)  Available bandwidth  Packet-pair technique – compute the transmission delay  What about cross-traffic?

4. Packet pair dispersion  The dispersion of the packet pair is the interval from the instant the last bit of the first packet is received at a certain path point to the instant the last bit of the second packet is received at that point.  When there is no cross traffic, all the bandwidth estimates are equal to the capacity  Post-narrow link - when the first probing packet is delayed more than the second  In heavy load conditions, the probability of CT packets interfering with the probing packets is large

5. Cross traffic interference  Capacity Mode (CM)  Sub-Capacity Dispersion Range (SCDR)  Post-Narrow Capacity Modes (PNCMs)

6. Packet train dispersion 1/2  As the length N of the train increases, the CM and PCNMs become weaker and the SCDR prevails.  As N increases, the dispersion becomes unimodal.  The range of the distribution, which is related to the measurement variance, decreases as N increases.  When N is sufficiently large, the center of the (unique) mode (a.k.a. Asymptotic Dispersion Rate) is independent on N.

7. Packet train dispersion 2/2

8. The size of probing packets  A large packet size leads to a wide time interval in which a CT packet can interfere with the packet pair.  When it is small, the formation of PNCMs becomes more likely and the CM becomes weaker.  The empirical conclusion from Internet experiments is that a packet size around 800 bytes leads to the stronger CM in heavily loaded paths.

9. Small vs large packet size

10. The “pathrate” methodology  Phase I: Packet pair probing.  use a large number of packet pair experiments to ‘uncover’ all the local modes of the bandwidth distribution.  Denote the sequence M of local modes in increasing order (one of these modes is the CM)  Phase II: Packet train probing.  Find Asymptotic Dispersion Rate and use heuristic rule to estimate the capacity mode

11. Heuristic evaluation  the resolution (bin width) has to be chosen based on a rough estimate of the path’s capacity.  the specified bin width is 1 Mbps for capacity paths 40 Mbps.

12. Conclusion  We studied the dispersion of packet pairs and packet trains, focusing on the effects of the cross traffic.  The insight gained can be applied to congestion control mechanisms, server selection algorithms and quality of service monitoring.

13. References  Constantinos Dovrolis, Parameswaran Ramanathan, David Moore: What do packet dispersion techniques measure? In INFOCOM 01.What do packet dispersion techniques measure?