Abstract
Tools for estimating end-to-end available bandwidth (AB) send out a train of packets and observe how inter-packet gaps change over a given network path. In ultra-high speed networks, the fine inter-packet gaps are fairly susceptible to noise introduced by transient queuing and bursty cross-traffic. Past work uses smoothing heuristics to alleviate the impact of noise, but at the cost of requiring large packet trains. In this paper, we consider a machine-learning approach for learning the AB from noisy inter-packet gaps. We conduct extensive experimental evaluations on a 10 Gbps testbed, and find that supervised learning can help realize ultra-high speed bandwidth estimation with more accuracy and smaller packet trains than the state of the art. Further, we find that when training is based on: (i) more bursty cross-traffic, (ii) extreme configurations of interrupt coalescence, a machine learning framework is fairly robust to the cross-traffic, NIC platform, and configuration of NIC parameters.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsNotes
- 1.
We focus on 10 Gbps speed in this paper, and use jumbo frames of MTU=9000B.
- 2.
- 3.
Probing range is given by: \(\frac{r_{N}}{r_{1}}-1\).
- 4.
Our evaluations revealed that models trained with ElasticNet and SVM result in considerable inaccuracy. For brevity, we don’t present their results.
- 5.
In our Python implementation with scikit-learn [22] library, we use its automatic parameter tuning feature for all ML methods, and use 5-fold cross-validation to validate our results.
- 6.
Note that replayed traffic retains the burstiness of original traffic aggregate, but does not retain responsiveness of individual TCP flows. However, the focus of this paper is to evaluate denoising techniques for accurate AB estimation —this metric is not impacted by the responsiveness of cross traffic, but only by its burstiness.
- 7.
Each weak model in RandomForest is learned on a different subset of training data. The final prediction is the average result of all models. AdaBoost and GradientBoost follow a boosting approach, where each model is built to emphasize the training instances that previous models do not handle well. The boosting methods are known to be more robust than RandomForest [25], when the data has few outliers.
- 8.
Since models are trained off-line, the training overhead is not of concern.
References
Dykes, S.G., et al.: An empirical evaluation of client-side server selection algorithms. In: INFOCOM 2000 (2000)
Aboobaker, N., Chanady, D., Gerla, M., Sanadidi, M.Y.: Streaming media congestion control using bandwidth estimation. In: Almeroth, K.C., Hasan, M. (eds.) MMNS 2002. LNCS, vol. 2496, pp. 89–100. Springer, Heidelberg (2002)
Konda, K.: RAPID: shrinking the congestion-control timescale. In: INFOCOM. IEEE (2009)
Jain, D.: Pathload: a measurement tool for end-to-end available bandwidth. In: PAM (2002)
Ribeiro, V., et al.: pathchirp: Efficient available bandwidth estimation for network paths. In: PAM, vol. 4 (2003)
Cabellos-Aparicio, A., et al.: A novel available bandwidth estimation and tracking algorithm. In: NOMS. IEEE (2008)
Shriram, A., Kaur, J.: Empirical evaluation of techniques for measuring available bandwidth. In: INFOCOM. IEEE (2007)
Kang, S.-R., Loguinov, D.: IMR-pathload: robust available bandwidth estimation under end-host interrupt delay. In: Claypool, M., Uhlig, S. (eds.) PAM 2008. LNCS, vol. 4979, pp. 172–181. Springer, Heidelberg (2008)
Kang, S.R., Loguinov, D.: Characterizing tight-link bandwidth of multi-hop paths using probing response curves. In: IWQoS. IEEE (2010)
Yin, Q., et al.: Can bandwidth estimation tackle noise at ultra-high speeds?. In: ICNP. IEEE (2014)
Strauss, J., et al.: A measurement study of available bandwidth estimation tools. In: The 3rd ACM SIGCOMM Conference on Internet Measurement (2003)
Lee, K.-S.: SoNIC: precise realtime software access and control of wired networks. In: NSDI (2013)
Rizzo, L.: netmap: A novel framework for fast packet I/O. In: USENIX Annual Technical Conference, pp. 101–112 (2012)
Prasad, R., Jain, M., Dovrolis, C.: Effects of interrupt coalescence on network measurements. In: Barakat, C., Pratt, I. (eds.) PAM 2004. LNCS, vol. 3015, pp. 247–256. Springer, Heidelberg (2004)
Dietterich, T.G.: Machine-learning research (1997)
Nguyen, T.T., et al.: A survey of techniques for internet traffic classification using machine learning. Commun. Surv. Tutor. 10(4), 56–76 (2008)
Zou, H., et al.: Regularization, variable selection via the elastic net. J. R. Stat. Soc.: Ser. B (Stat. Methodol.) 67(2), 301–320 (2005)
Liaw, A., et al.: Classification and regression by randomforest. R News 2(3), 18–22 (2002)
Freund, Y., et al.: A decision-theoretic generalization of on-line learning and an application to boosting. J. Comput. Syst. Sci. 55(1), 119–139 (1997)
Friedman, J.H.: Greedy function approximation: a gradient boosting machine. Ann. Stat. 29(5), 1189–1232 (2001)
Cortes, C., et al.: Support-vector networks. Mach. Learn. 20(3), 273–297 (1995)
Pedrogosa, F., et al.: Scikit-learn: machine learning in python. J. Mach. Learn. Res. 12, 2825–2830 (2011)
Barford, P., Crovella, M.: Generating representative web workloads for network and server performance evaluation. ACM SIGMETRICS Perform. Eval. Rev. 26(1), 151–160 (1998)
Turner, A.A., Bing, M.: Tcpreplay (2005)
Dietterich, T.: An experimental comparison of three methods for constructing ensembles of decision trees: bagging, boosting, and randomization. Mach. Learn. 40(2), 139–157 (2000)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2016 Springer International Publishing Switzerland
About this paper
Cite this paper
Yin, Q., Kaur, J. (2016). Can Machine Learning Benefit Bandwidth Estimation at Ultra-high Speeds?. In: Karagiannis, T., Dimitropoulos, X. (eds) Passive and Active Measurement. PAM 2016. Lecture Notes in Computer Science(), vol 9631. Springer, Cham. https://doi.org/10.1007/978-3-319-30505-9_30
Download citation
DOI: https://doi.org/10.1007/978-3-319-30505-9_30
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-30504-2
Online ISBN: 978-3-319-30505-9
eBook Packages: Computer ScienceComputer Science (R0)