draft-duffield-ippm-burst-loss-metrics-01.txt

1. draft-duffield-ippm-burst-loss-metrics-01.txt Nick Duffield, Al Morton, AT&TJoel Sommers, Colgate University IETF 76, Hiroshima, Japan 11/10/2009

2. Agenda History of the draft One page summary of draft Mailing list comments and discussion Related activity Conclusions

3. History of draft-duffield-ippm-burst-loss-metrics • Aim: standardize measurement of loss episodes [SBDR08] • Initial presentations IETF 72, 73 • -00 individual draft published prior to IETF 74 • IPR disclosures for -00 draft completed April 2009 • -01 draft published July 2009 • Open question: should draft be adopted as WG item? • Some comments and questions on draft on the IPPM mailing list • Thanks for comments; more please!

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 loss 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 • Probing: bi-packet probes, sent as discrete Poisson stream • Analysis: metrics depend only on frequencies 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. Mailing list comments and discussion Should metrics be loss episodes average or general burstiness? What is the relation to Gilbert model? Should metrics be time based or count based? Need for clarification of role of selection function

6. Metrics: Loss episode averages or burstiness? Structure of loss episodes is more complex that average length Multipacket statistics? (Prob[episode has n packets], n = 1,2,3,…) Correlations between lengths of episodes, gaps between episodes? Questions/Issues: What is added utility of multipacket statistics over averages? Metric statistical accuracy decreases with number of packets n Authors’ Recommendation Retain only loss episode averages (simple extension of RFC 2680) Defer multipacket loss statistics as separate WG item if interest NB: averaging metrics do not need to sample full loss episodes

7. What is relation to Gilbert model? • In parametric terms, the Gilbert model is more complex • Gilbert model has 4 parameters: Good/Bad state lifetimes/loss rate • Two independent loss episode metrics (average duration, frequency) • Metrics are purely empirical, interpreted independent of model • Metrics do not aim to estimate parameters of any model • Authors’ Recommendation: expand draft with applicability section

8. Time based or count based episode metrics? • Some well known burstiness metric are based on packet counts • IDC: index of dispersion on counts • Loss episode metrics based on time (average duration etc) • Easier to compare directly with application requirements • Probe rate and traffic rate generally different • Authors’ Recommendation: • Retain time-basis

9. What is the role of selection function? • Selection function is a general formulation of a way to specify which packet are used for probing (see RFC 3393) • Examples: • Specifying how discrete Poisson b-packet probes are to be selected • Potential use to specify selection mechanism for background traffic to be co-opted as probes. • Authors’ Recommendation: • Expand explanation of selection function in draft.

10. Related Activity • ITU Activity • contributed to ITU-T SG 12 Question 17 on packet performance. • independent implementation of same loss episode metrics • Special case: unsampled counts of 4 bi-packet outcomes

11. Authors’ Conclusions Mailing list discussion has been helpful and constructive; thanks! Points raised appear to request clarifications and elaboration, rather than raising fundamental objections to metrics or methods Authors will update accordingly in next draft version