V. Padmanabhan, L. Qiu, and H. Wang. Server-based inference of Internet performance. In IEEE INFOCOM, Apr. 2003.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....detailed data for the AS topology are collected since 1997 [22] No router level data that span that many years exist. Third, bottlenecks in the Internet are usually either at the access links, or at the connections between ASes (on the contrary, links inside the ASes are usually overprovisioned) [45]. The construction of trees with preferential connectivity and 3 regular expanders is straightforward. Our 3 regular expanders of size n are random graphs with n nodes, where Table 1: Congestion of the AS topology Year Nodes Links Shortest Hop Shortest Hop with Policies 1997 3055 5238 559653 ....

V. Padmanabhan, L. Qiu, and H. Wang. Server-based inference of internet performance. In Proc. Infocom. IEEE, 2003. To appear.

....OF END TO END LOSS RATE We analyzed the end to end loss rate information derived from traffic traces gathered at the microsoft.com site. Due to space limitations, we only present a sketch of our experimental methodology and our key findings here. The technical report version of this paper [15] includes a detailed description of our experimental setup, methodology, and results. The traces were gathered by running the tcpdump tool on a machine connected to the replication port on a Cisco Catalyst 6509 switch. The packet sniffer was thus able to listen on all communication between the ....

....tend to have low loss rates. Zhang et al. 20] reported that the loss rate remains operationally stable on the time scale of an hour. Our temporal locality analysis based on the microsoft.com traces indicates a stability duration ranging from several minutes to tens of minutes (Section III and [15]) So it is reasonable to perform inference based on end to end packet loss information gathered over such time scales. Second, even when the loss rate of each link is constant, it may not be possible to definitively identify the loss rate of each link. Given M clients and N links, we have M ....

V. N. Padmanabhan, L. Qiu, and H. J. Wang. Server-based Inference of Internet Performance. Technical Report MSR-TR-2002-39, Microsoft Research, May 2002.

....tomography: Random Sampling, Linear Optimization, and Bayesian Inference using Gibbs Sampling. We have evaluated these techniques using simulations and traces gathered at a busy Web server. In this paper, we focus on the Gibbs Sampling technique; more information on the three techniques appears in [3]. II. BAYESIAN INFERENCE USING GIBBS SAMPLING We model passive network tomography as a Bayesian inference problem. We first present some background information. A. Background Let D denote the observed data and denote the (unknown) model parameters. In the context of network tomography, D ....

....a few hundred iterations) we obtain samples from the desired distribution, P (l L jD) We use these samples to determine which links are likely to be lossy. III. PERFORMANCE EVALUATION We evaluate the inference technique using both simulations and real packet traces. Detailed results appear in [3]. A. Simulation Results The main advantage of simulation is that the true link loss rates are known, so validating the inferences of network tomography is easy. The simulation experiments are performed on topologies of different sizes using multiple link loss models. For each topology, we set ....

V. N. Padmanabhan, L. Qiu, and H. J. Wang. Server-based inference of internet performance. In Microsoft Research Technical Report MSR-TR2002 -39, May 2002.

....tomography: Random Sampling, Linear Optimization, and Bayesian Inference using Gibbs Sampling. We have evaluated these techniques using simulations and traces gathered at a busy Web server. In this paper, we focus on the Gibbs Sampling technique; more information on the three techniques appears in [3]. II. BAYESIAN INFERENCE USING GIBBS SAMPLING We model passive network tomography as a Bayesian inference problem. We first present some background information. A. Background Let D denote the observed data and # denote the (unknown) model parameters. In the context of network tomography, D ....

....lasts a few hundred iterations) we obtain samples from the desired distribution, D) We use these samples to determine which links are likely to be lossy. III. PERFORMANCE EVALUATION We evaluate the inference technique using both simulations and real packet traces. Detailed results appear in [3]. A. Simulation Results The main advantage of simulation is that the true link loss rates are known, so validating the inferences of network tomography is easy. The simulation experiments are performed on topologies of different sizes using multiple link loss models. For each topology, we set ....

V. N. Padmanabhan, L. Qiu, and H. J. Wang. Server-based inference of internet performance. In Microsoft Research Technical Report MSR-TR2002 -39, May 2002.

No context found.

V. Padmanabhan, L. Qiu, and H. Wang. Server-based inference of Internet performance. In IEEE INFOCOM, Apr. 2003.

No context found.

V. Padmanabhan, L. Qiu, and H. Wang. Server-based inference of Internet performance. In IEEE INFOCOM, Apr. 2003.

No context found.

V. N. Padmanabhan, L. Qiu, and H. Wang. Server-based inference of internet performance. In IEEE INFOCOM'03, San Francisco, CA, USA, April 2003.

No context found.

V. Padmanabhan, L. Qiu, and H. Wang. Server-based inference of Internet performance. In IEEE INFOCOM, Apr. 2003.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC