draft-ietf-ippm-loss-episode-metrics-00 Loss Episode Metrics for IPPM

1. draft-ietf-ippm-loss-episode-metrics-00Loss Episode Metrics for IPPM Nick Duffield, Al Morton, AT&TJoel Sommers, Colgate University IETF 79, Beijing, 11/8/2010

2. Agenda Changes since IETF 77 One page summary of draft Q&A from last IETF Conclusion

3. Changes since IETF 77,78 • Draft adopted as WG item • Previous individual draft draft-duffield-ippm-burst-loss-metrics • IPR disclosure • Supersedes IPR disclosure for previous individual draft • https://datatracker.ietf.org/ipr/1354/

4. X X X X X X X X A one page summary of the draft • Frequent small glitches vs. local burst (at same average loss rate) (0,0) (0,1) (1,1) (0,0) • Fact: packets in a flow are not generally lost independently • Motivation: metrics of temporal structure of packet loss • Target use: SLAs, application requirements (e.g VoIP) • Object of study: loss episodes (of consecutively loss packets) • Metrics: average duration and frequency of loss episodes • Methodology: bi-packet probes, sent as discrete Poisson stream • Analysis: metrics depend only on frequencies of probe outcomes • 4 possible outcomes (0,0), (0,1), (1,1), (1,0) where 1 = lost, 0 = not lost • Summary: extension of RFC 2680 to case of correlated loss

5. Relation to Gilbert Model • Metric maps to 2-state Gilbert model • Draft seeks to provide simplest extension of RFC2680 that can capture dependent loss without reference to a model • Possible work for future WG drafts? • Standardize specific measurement methodology for estimating the parameters of the 4-state Gilbert model: • It should be coordinated with related work, such as RFC 3611 (RTCP-XR) and ITU-T Rec. G.1020

6. Sampling Packet Pairs vs. Capture All Packets • Draft Approach: Sample packet pairs. • Sample transitions between runs of lost/not_lost packets • These are sufficient to compute draft metrics • Don’t need to capture all packets in a loss episode • Relation to other measurement methodologies • Passive measurement of existing stream • Yes, you can measure all packets (work on Stream-Repair metrics in SG12) • Equivalent to setting launch probability q = 1 • Metrics are identical with computation based on runs of successive losses. • Active probing • ANY probing scheme samples loss at a sequence of instants • Example RFC 2680 • Packet pairs are more efficient at sampling the transitions • For a given accuracy, less probing bandwidth required than for stream of single packets

7. Conclusion Please read the draft and comment Please post any questions from this WG session to the mailing list