H H Ch 1 Ch C Hung-Hsuan Chen 1 , Cheng-C M Hung Hsuan Chen , Cheng C M 1 P l i St t Ui it 2 St M 1 Pennsylvania State University, 2 Sto Pennsylvania State University, Sto Lab Lab M ti ti M ti ti M ti ti M ti ti Mthdl Mthdl Mthdl Mthdl Motivation Motivation Motivation Motivation Methodology Methodology Methodology Methodology Motivation Motivation Motivation Motivation Methodology Methodology Methodology Methodology A f li i ’ k b E bli h h l i hi b Assessment of applications’ network robustness Establish the relationship betwee Assessment of applications network robustness Establish the relationship betwee fh k h Network of the network path Application Delay Loss Rate Network of the network path Application Delay Loss Rate Rob stness The departure rate can be divided Robustness The departure rate can be divided ? Application specific: “ideal” sce 1 Good Poor ? Application‐specific: “ideal” sce 1 Good Poor ? 2 Poor Good ? 2 Poor Good ? ? hi h li i h b k b ? Which application has better network robustness? Which application has better network robustness? h i h i h i h i Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis Hypothesis U th b tj d f t k li ti ’ f Users are the best judge of a network application’s performance Risk score: a magnifier summar Risk score: a magnifier summar impairment comparable amon Poor Network Quality impairment, comparable amon Poor Network Quality (Risk score ↑ network robus (Risk score ↑ network robus ff affects Unstable Game Play affects Unstable Game Play f db Verified by Verified by real-life traces real life traces in our previous in our previous Less Fun works Less Fun works Sh G Pl Ti Shorter Game Play Time Shorter Game Play Time Adopting Gaussian mixture mode Adopting Gaussian‐mixture mode scenarios (number of mixture com scenarios (number of mixture com Contribution Contribution Contribution Contribution Contribution Contribution Contribution Contribution Contribution Contribution Contribution Contribution A framework for quantifying an application’s network robustness A framework for quantifying an application s network robustness D t ti th ff ti l lif dt t Demonstrating the effectiveness on real‐life data traces Ch T 2 d K T Ch 3 Chun Tu 2 , and Kuan-Ta Chen 3 Chun Tu , and Kuan Ta Chen B k U i it 3 A d i Si i ony Brook University 3 Academia Sinica ony Brook University Academia Sinica N t k R b t Id N t k R b t Id N t k R b t Id N t k R b t Id Network Robustness Index Network Robustness Index Network Robustness Index Network Robustness Index Network Robustness Index Network Robustness Index Network Robustness Index Network Robustness Index 1 d dh li 1 en user departure rates and the quality = 1 (NRI) Index Robustness Network en user departure rates and the quality ∫ = ) ( score risk (NRI) Index Robustness Network ∫ ) ( score risk ∫ d into two components: d into two components: 1 enario no network impairment 1 enario, no network impairment = p ∫ ∑ ) ) exp( ( dx dx x ∫ ∑ β ) ,..., ) exp( ( 1 p i i dx dx x ∫ ∑ β 1 ,..., 1 1 p x x i i p ∫ ∑ 1 1 i p = Ri k S Diff Ri k S Diff Ri k S Diff Ri k S Diff Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences Risk Score Differences rizing the impact of network rizing the impact of network g different applications g different applications stness ↓) stness ↓) Most users in our traces experienced network scenarios located to the Most users in our traces experienced network scenarios located to the left of the equivalence line where the risk scores of SZ is lower than the left of the equivalence line where the risk scores of SZ is lower than the hypothetical game hypothetical game In most scenarios risk scores of SZ is lower than the hypothetical In most scenarios, risk scores of SZ is lower than the hypothetical game el to capture the target network el to capture the target network mponents 2) K l K l K l K l mponents: 2) Key Result Key Result Key Result Key Result Key Result Key Result Key Result Key Result Key Result Key Result Key Result Key Result f NRI of SZ: 6 58 NRI of SZ: 6.58 NRI fh h h i l 4 07 NRI of the hypothetical game: 4 07 NRI of the hypothetical game: 4.07 O SZ’ t k f i b tt On average SZ’s network performance is better On average, SZ s network performance is better Multimedia Networking and Systems Lab Multimedia Networking and Systems Lab Instit te of Information Science Academia Sinica Institute of Information Science, Academia Sinica Institute of Information Science, Academia Sinica http://mmnet iis sinica edu tw http://mmnet.iis.sinica.edu.tw