Academia.eduAcademia.edu

Figure 1 – uploaded by Constandinos Mavromoustakis

See full PDFdownloadDownload figure
Fig. 1. Network Architecture for Optimal Multimedia Services Provision

Figure 1 Network Architecture for Optimal Multimedia Services Provision

Related Figures (10)

Fig. 2. Media Distribution Middleware Internal Architectur component gathers the monitoring data that comes from the MQM and uses it as input to the Media Traffic Forecast, generating the prediction for the traffic in the network. The outcome of the forecast is used as an input to the Resource allocator/scheduler that takes the decision for the optimal delivery methods, feeding the MSM component with the recommendations, on which server should stream the requested media. At the same time, the Bandwidth Allocation Optimizer
Fig. 2. Media Distribution Middleware Internal Architectur component gathers the monitoring data that comes from the MQM and uses it as input to the Media Traffic Forecast, generating the prediction for the traffic in the network. The outcome of the forecast is used as an input to the Resource allocator/scheduler that takes the decision for the optimal delivery methods, feeding the MSM component with the recommendations, on which server should stream the requested media. At the same time, the Bandwidth Allocation Optimizer
Fig. 3. Proposed Online Resource Prediction System be used by the Resource allocator/scheduler to decide how to serve the estimated future need.
Fig. 3. Proposed Online Resource Prediction System be used by the Resource allocator/scheduler to decide how to serve the estimated future need.
where D=[D,], x=1,..,n represents the combined delivery method for each of the n VoD channels, Dejoyg=[Dx], x=1,...n corresponds to the delivery over the cloud, Depy=[Dx], x=1,..,.n to the delivery with use of CDN and Dyye¢c=[D,], x=1,..,n to the use of P2P network between EHG devices. In the formula W=[W,yy], X=I,...n, y=l,.,m, where W,y represents the proportion of video channel x’s requests directed to Data Center(DC) y. A=[a,] and B=[b,], x=1,...,n, where the variables can only take the values 0 and | to enable or disable the use of Cloud and P2P delivery respectively. In case of enabling P2P delivery, for a specific VoD, EHGs devices implement proper P2P algorithms to take the decisions about the load distribution. The corresponding algorithm as implemented in the MDM is presented below. It can be observed that the higher value can be taken among the current and the predicted bandwidth need. If it is below a preset limit (based on administrative high level decisions and network status), only the Cloud delivery will be used. If it is above the limit, CDNs will be used and the number of CDNs used is increasing according the demand. After the top limit is reached, a P2P delivery method is exploited. This algorithm provides the advantage that the data is not distributed before the actual need. In the case of a VoD with low customers demand, the CDNs will not been used for its distribution. On the other hand, if a popular VoD is requested, an early prediction will occur, while the number of CDNs, distributing the VoD will rapidly increase based on the demand. Finally, the P2P delivery method will be used, only when needed, while at the time that this happens, the number of the users already possessing the specific video will be satisfactory with those users, acting as seeders to distribute the VoD to the others.
where D=[D,], x=1,..,n represents the combined delivery method for each of the n VoD channels, Dejoyg=[Dx], x=1,...n corresponds to the delivery over the cloud, Depy=[Dx], x=1,..,.n to the delivery with use of CDN and Dyye¢c=[D,], x=1,..,n to the use of P2P network between EHG devices. In the formula W=[W,yy], X=I,...n, y=l,.,m, where W,y represents the proportion of video channel x’s requests directed to Data Center(DC) y. A=[a,] and B=[b,], x=1,...,n, where the variables can only take the values 0 and | to enable or disable the use of Cloud and P2P delivery respectively. In case of enabling P2P delivery, for a specific VoD, EHGs devices implement proper P2P algorithms to take the decisions about the load distribution. The corresponding algorithm as implemented in the MDM is presented below. It can be observed that the higher value can be taken among the current and the predicted bandwidth need. If it is below a preset limit (based on administrative high level decisions and network status), only the Cloud delivery will be used. If it is above the limit, CDNs will be used and the number of CDNs used is increasing according the demand. After the top limit is reached, a P2P delivery method is exploited. This algorithm provides the advantage that the data is not distributed before the actual need. In the case of a VoD with low customers demand, the CDNs will not been used for its distribution. On the other hand, if a popular VoD is requested, an early prediction will occur, while the number of CDNs, distributing the VoD will rapidly increase based on the demand. Finally, the P2P delivery method will be used, only when needed, while at the time that this happens, the number of the users already possessing the specific video will be satisfactory with those users, acting as seeders to distribute the VoD to the others.
IV. PERFORMANCE EVALUATION ANALYSIS, EXPERIMENTAL RESULTS AND DISCUSSION DC should handle. Then, it finds out which column represents the specific VoD or assigns a new column for a new VoD. Finally, it selects which DCs will be used, starting with those that have more zero values in their row, meaning that they do not serve many VoD channels.
IV. PERFORMANCE EVALUATION ANALYSIS, EXPERIMENTAL RESULTS AND DISCUSSION DC should handle. Then, it finds out which column represents the specific VoD or assigns a new column for a new VoD. Finally, it selects which DCs will be used, starting with those that have more zero values in their row, meaning that they do not serve many VoD channels.
This section demonstrates the effectiveness of the resource prediction engine as a subsystem, by presenting experimental results, in terms of the outcome of the engine compared to the actual values measured, as well as the effectiveness of the whole proposed system, by executing simulations of the usage scenarios. For the evaluation of the forecast algorithms, the monitoring data of the bandwidth usage for serving the need of a specific VoD, was utilized. The collected measurements were for a total of 30 minutes with a period of 5 seconds, but to avoid periodicity of data, the mean value per minute was used. The 80% (24 minutes) of the data was exploited to feed the prediction engine, while the rest of the data was used for the evaluation, through a comparison between the predicted and the actual value as shown in Fig. 4. The important part of the graph is after the 24 first minutes, where it is clearly depicted that the measured values remain very close to the predicted ones. Fig. 4. Measured vs Predicted value for the bandwidth usage for VoD
This section demonstrates the effectiveness of the resource prediction engine as a subsystem, by presenting experimental results, in terms of the outcome of the engine compared to the actual values measured, as well as the effectiveness of the whole proposed system, by executing simulations of the usage scenarios. For the evaluation of the forecast algorithms, the monitoring data of the bandwidth usage for serving the need of a specific VoD, was utilized. The collected measurements were for a total of 30 minutes with a period of 5 seconds, but to avoid periodicity of data, the mean value per minute was used. The 80% (24 minutes) of the data was exploited to feed the prediction engine, while the rest of the data was used for the evaluation, through a comparison between the predicted and the actual value as shown in Fig. 4. The important part of the graph is after the 24 first minutes, where it is clearly depicted that the measured values remain very close to the predicted ones. Fig. 4. Measured vs Predicted value for the bandwidth usage for VoD
Fig. 7. Total delay (usec) with number of simultaneous transmissions Fig. 6. Number of participating nodes with community streaming factor
Fig. 7. Total delay (usec) with number of simultaneous transmissions Fig. 6. Number of participating nodes with community streaming factor
The rest of the section presents experimental simulation results for evaluation of the performance and the offered reliability in streaming activities, offered by the proposed system. More specifically, Fig. 6 shows that the number of the participating nodes is increasing, when MDM-enhancing broadcasting is used, instead of a generic broadcasting. This indicates the enhancement that has been done by the MDM in the broadcasting process, whereas the Community Streaming factor W, as introduced in [19], indicates the level of robustness in receiving neighboring feedback during the process of streaming. The total delay time with the number of simultaneous transmissions is shown in Fig. 7. The total measured delay is significantly reduced in the presence of MHGC (P2P delivery), whereas the utilization of the existing infrastructure increases the overall delays when multiple transmissions take place.
The rest of the section presents experimental simulation results for evaluation of the performance and the offered reliability in streaming activities, offered by the proposed system. More specifically, Fig. 6 shows that the number of the participating nodes is increasing, when MDM-enhancing broadcasting is used, instead of a generic broadcasting. This indicates the enhancement that has been done by the MDM in the broadcasting process, whereas the Community Streaming factor W, as introduced in [19], indicates the level of robustness in receiving neighboring feedback during the process of streaming. The total delay time with the number of simultaneous transmissions is shown in Fig. 7. The total measured delay is significantly reduced in the presence of MHGC (P2P delivery), whereas the utilization of the existing infrastructure increases the overall delays when multiple transmissions take place.
Fig. 5. Test scenarios for bandwidth usage of VoD prediction
Fig. 5. Test scenarios for bandwidth usage of VoD prediction
Fig. 8. CCDF sharing reliability among device with download time (msec) for requests <20 MB fading channels with noise have low throughput exhibition especially when the number of requests per second increases.
Fig. 8. CCDF sharing reliability among device with download time (msec) for requests <20 MB fading channels with noise have low throughput exhibition especially when the number of requests per second increases.
Related topics:
Computer Science
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved