ABSTRACT
Packet delay and loss are two fundamental measures of performance. Using active probing to measure delay and loss typically involves sending Poisson probes, on the basis of the PASTA property (Poisson Arrivals See Time Averages), which ensures that Poisson probing yields unbiased estimates. Recent work, however, has questioned the utility of PASTA for probing and shown that, for delay measurements, i) a wide variety of processes other than Poisson can be used to probe with zero bias and ii) Poisson probing does not necessarily minimize the variance of delay estimates.
In this paper, we determine optimal probing processes that minimize the mean-square error of measurement estimates for both delay and loss. Our contributions are twofold. First, we show that a family of probing processes, specifically Gamma renewal probing processes, has optimal properties in terms of bias and variance. The optimality result is general, and only assumes that the target process we seek to optimally measure via probing, such as a loss or delay process, has a convex auto-covariance function. Second, we use empirical datasets to demonstrate the applicability of our results in practice, specifically to show that the convexity condition holds true and that Gamma probing is indeed superior to Poisson probing. Together, these results lead to explicit guidelines on designing the best probe streams for both delay and loss estimation.
- F. Baccelli and P. Bremaud. Elements of Queueing Theory. Springer Verlag, Applications of Mathematics, second edition, 2003.Google Scholar
- F. Baccelli, S. Machiraju, D. Veitch, and J. Bolot. The Role of PASTA in Network Measurement. Computer Communication Review, Proceedings of ACM Sigcomm 2006, 36(4):231--242, 11-15 Sep 2006. Google ScholarDigital Library
- F. Baccelli, S. Machiraju, D. Veitch, and J. Bolot. Loss Measurement via Probing. Sprint ATL Technical Report No. ATL-020124, February 2007.Google Scholar
- D. Daley and D. Vere-Jones. An Introduction to the Theory of Point Processes. Springer-Verlag, 1988.Google Scholar
- D. Gaver and P. Lewis. First-order autoregressive gamma sequences and point processes. Adv. Appl. Prob., 12:727--745, 1980.Google ScholarCross Ref
- L. Gradshteyn and L. Ryzhik. Table of Integrals, Series and Products. Academic Press, sixth edition, 2000.Google Scholar
- N. Hohn, D. Veitch, K. Papagiannaki and C. Diot. Bridging router performance and queuing theory. In Proc. of ACM SIGMETRICS'04. Google ScholarDigital Library
- I. I. P. M. (IPPM). http://www.ietf.org/html.charters/ippm-charter.html, 2005.Google Scholar
- S. Machiraju, D. Veitch, F. Baccelli, and J. Bolot. Adding Definition to Active Probing. In ACM Computer Communication Review, April 2007. Google ScholarDigital Library
- M. Tariq, A. Dhamdhere, C. Dovrolis, and M. Ammar. Poisson versus Periodic Path Probing (or, Does PASTA Matter)? In ACM Internet Measurement Conf., pages 119--124, Berkeley, CA, Oct 2005. Google ScholarDigital Library
- T. Ott. The covariance function of the virtual waiting time process in an M/G/1 queue. Adv. App. Prob., 9, 1997.Google Scholar
- D. Papagiannaki, S. Moon, C. Fraleigh, P. Thiran, F. Tobagi and C. Diot. Analysis of Measured Single-hop Delay from an Operational Backbone Network In Proceedings of IEEE Infocom, June 2002.Google Scholar
- D. Papagiannaki, D. Veitch and N. Hohn. Origins of Microcongestion in an Access Router. In Proceedings of Passive and Active Measurement Conference., 2004.Google ScholarCross Ref
- V. Paxson. End-to-end routing behavior in the internet. In Proceedings of ACM SIGCOMM, 1997. Google ScholarDigital Library
- V. Paxson. End-to-end Internet packet dynamics. IEEE/ACM Transactions on Networking, 7(3):277--292, 1999. Google ScholarDigital Library
- K. Petersen. Ergodic Theory. Cambridge University Press, Cambridge England, 1983.Google Scholar
- M. Roughan. A Comparison of Poisson and Uniform Sampling for Active Measurements. IEEE Journal on Selected Areas in Communication, 24(12):2299--2312, Dec 2006. Google ScholarCross Ref
- J. Sommers, P. Barford, N. Duffield, and A. Ron. Improving accuracy in end-to-end packet loss measurement. In ACM SIGCOMM'05, pages 157--168, 2005. Google ScholarDigital Library
- L. Takács. Introduction to the Theory of Queues. Oxford University Press, New York, 1962.Google Scholar
- The Network Simulator - ns-2, 2004. http://www.isi.edu/nsnam/ns/.Google Scholar
- R. Wolff. Poisson Arrivals see Time Averages. Operations Research, 3(2):223--231, 1982.Google ScholarDigital Library
- Y. Zhang, N. Duffield, and V. Paxson. On the Constancy of Internet Path Properties. In Proceedings of the Internet Measurement Workshop, 2001. Google ScholarDigital Library
Index Terms
- On optimal probing for delay and loss measurement
Recommendations
The role of PASTA in network measurement
Proceedings of the 2006 conference on Applications, technologies, architectures, and protocols for computer communicationsPoisson Arrivals See Time Averages (PASTA) is a well known property applicable to many stochastic systems. In active probing, PASTA is invoked to justify the sending of probe packets (or trains) at Poisson times in a variety of contexts. However, due to ...
Determining an adequate probe separation for estimating the arrival rate in an M/D/1 queue using single-packet probing
We present a technique to estimate the arrival rate from delay measurements, acquired using single-packet probing. With practical applications in mind, we investigate a lower bound on the value of probe separation, to obtain a satisfactory estimate in a ...
Convexity Results for the Erlang Delay and Loss Formulae When the Server Utilization Is Held Constant
This paper proves a long-standing conjecture regarding the optimal design of the M/M/s queue. The classical Erlang delay formula is shown to be a convex function of the number of servers when the server utilization is held constant. This means that when ...
Comments