Academia.eduAcademia.edu

Figure 5 – uploaded by Javier Morgade

See full PDFdownloadDownload figure
T; is the variable that really changes across every experiment depending on the cost-effective strategy, leading to different utilization rates for each available CDN. Then, we employ 7; as the main metric for comparing the OPEX achieved by the different strategies.

Table 5 T; is the variable that really changes across every experiment depending on the cost-effective strategy, leading to different utilization rates for each available CDN. Then, we employ 7; as the main metric for comparing the OPEX achieved by the different strategies.

Related Figures (13)

Fig. 2. Sequence diagram of the INFLOW solution for video delivery. Fig. 1. General scenario of the proposed solution.
Fig. 2. Sequence diagram of the INFLOW solution for video delivery. Fig. 1. General scenario of the proposed solution.
Fig. 3. Testbed setup: configurations for dataset creation (a) and for INFLOW enabled delivery (b).
Fig. 3. Testbed setup: configurations for dataset creation (a) and for INFLOW enabled delivery (b).
Fig. 4. Bandwidth and latency autocorrelation.
Fig. 4. Bandwidth and latency autocorrelation.
Fig. 5. Bandwidth and latency prediction: validation of the LSTM model.
Fig. 5. Bandwidth and latency prediction: validation of the LSTM model.
Fig. 6. Bandwidth and latency prediction: testing of the trained LSTM model.
Fig. 6. Bandwidth and latency prediction: testing of the trained LSTM model.
AVERAGE VALUE AND STANDARD DEVIATION OF THE MEASURED BANDWIDTH BY THE PLAYERS TABLE III AVERAGE VALUE AND STANDARD DEVIATION OF THE SELECTED BITRATE BY THE PLAYERS
AVERAGE VALUE AND STANDARD DEVIATION OF THE MEASURED BANDWIDTH BY THE PLAYERS TABLE III AVERAGE VALUE AND STANDARD DEVIATION OF THE SELECTED BITRATE BY THE PLAYERS
CDN1 but underperforms at CDN2 and CDN3. This is because INFLOW aims to reduce the number of employed CDNs and the OPEX at any time, while guaranteeing at least 4 Mbps for the measured network bandwidth. Thus, video sessions to CDN2 or CDN3 can be retired, therefore saving the CDN OPEX. The other multi-CDN strategies do not reduce CDN usage, i.e., when the player is connected to a CDN, the con- nection is maintained until the session expires. In terms of fairness, INFLOW outperforms the other multi-CDN strategies as the average representation bitrates of each CDN are close to each other. Video players connected to CDN2 present the best representation bitrate (2.44 Mbps), which is +24% higher than those of video players connected to CDN1. Moreover, standard deviation for measured bandwidth and representation bitrate achieved with INFLOW also demonstrates that it is the fairest solution since the players experience almost the same variation independently of the CDN. Standard deviation for CDN2 is +58% higher than CDN1 if we consider measured bandwidth and +27% if we consider representation bitrate. ES and PS strategies limit to 33 (100 players / 3 CDNs) the number of players connected to CDN1. Both strategies pro- duce similar results for the different CDNs. For CDN1, the average measured bandwidth of ES and PS, when compared to single CDN strategy, improves +4.7% and +14.1%, respec- tively. These results mean a higher bitrate selection, +6.6% and +17.4%, respectively. Moreover, the selected bitrates are considerably higher for CDN2 and CDN3 as they provide a higher performance. These CDNs provide more network resources serving higher bandwidths for video players. The measured results range between 15.50 Mbps (ES) and 21.84 (PS) for CDN2 and 10.65Mbps (PS) and 11.42 (ES) for CDN3. Thus, distributing video players across the available CDNs by just considering the number of players per CDN (33 players) is not fair as video players connected to CDN2 and CDN3 can select higher representation bitrates. Video players select a representation bitrate up to +160% higher if connected to CDN2 and +168% higher if connected to CDN3.
CDN1 but underperforms at CDN2 and CDN3. This is because INFLOW aims to reduce the number of employed CDNs and the OPEX at any time, while guaranteeing at least 4 Mbps for the measured network bandwidth. Thus, video sessions to CDN2 or CDN3 can be retired, therefore saving the CDN OPEX. The other multi-CDN strategies do not reduce CDN usage, i.e., when the player is connected to a CDN, the con- nection is maintained until the session expires. In terms of fairness, INFLOW outperforms the other multi-CDN strategies as the average representation bitrates of each CDN are close to each other. Video players connected to CDN2 present the best representation bitrate (2.44 Mbps), which is +24% higher than those of video players connected to CDN1. Moreover, standard deviation for measured bandwidth and representation bitrate achieved with INFLOW also demonstrates that it is the fairest solution since the players experience almost the same variation independently of the CDN. Standard deviation for CDN2 is +58% higher than CDN1 if we consider measured bandwidth and +27% if we consider representation bitrate. ES and PS strategies limit to 33 (100 players / 3 CDNs) the number of players connected to CDN1. Both strategies pro- duce similar results for the different CDNs. For CDN1, the average measured bandwidth of ES and PS, when compared to single CDN strategy, improves +4.7% and +14.1%, respec- tively. These results mean a higher bitrate selection, +6.6% and +17.4%, respectively. Moreover, the selected bitrates are considerably higher for CDN2 and CDN3 as they provide a higher performance. These CDNs provide more network resources serving higher bandwidths for video players. The measured results range between 15.50 Mbps (ES) and 21.84 (PS) for CDN2 and 10.65Mbps (PS) and 11.42 (ES) for CDN3. Thus, distributing video players across the available CDNs by just considering the number of players per CDN (33 players) is not fair as video players connected to CDN2 and CDN3 can select higher representation bitrates. Video players select a representation bitrate up to +160% higher if connected to CDN2 and +168% higher if connected to CDN3.
AVERAGE VALUE AND STANDARD DEVIATION OF THE MEASURED LATENCY BY THE PLAYERS TABLE II
AVERAGE VALUE AND STANDARD DEVIATION OF THE MEASURED LATENCY BY THE PLAYERS TABLE II
Fig. 7. Distribution of the players among the available CDNs: single CDN (a), equal selection (b), progressive selection, (c) and INFLOW selection (d) strategies.
Fig. 7. Distribution of the players among the available CDNs: single CDN (a), equal selection (b), progressive selection, (c) and INFLOW selection (d) strategies.
Angel Martin received the Engineering degree from University Carlos III in 2003, and the Ph.D. degree from UPV/EHU in 2018. He is with the Department of Digital Media, Vicomtech, where he is working with multimedia services and 5G infrastructures projects. He developed in Prodys an standard MPEG-4 AVC/H.264 codec for DSP (2003-2005). He worked in media streaming and encoding research (2005-2008) with Telefonica. He worked in the fields of smart environments and ubiq- uitous and pervasive computing (2008-2010) with Innovalia.
Angel Martin received the Engineering degree from University Carlos III in 2003, and the Ph.D. degree from UPV/EHU in 2018. He is with the Department of Digital Media, Vicomtech, where he is working with multimedia services and 5G infrastructures projects. He developed in Prodys an standard MPEG-4 AVC/H.264 codec for DSP (2003-2005). He worked in media streaming and encoding research (2005-2008) with Telefonica. He worked in the fields of smart environments and ubiq- uitous and pervasive computing (2008-2010) with Innovalia.
Related topics:
Computer ScienceContent Delivery NetworkStreaming MediaElectrical and Electronic EngineeringOperational expenditure (OpEx)MPEG-DASH
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved