CORDIS - EU research results
CORDIS

P2P middleware for the deployment of an innovative business model for the provision of a QoS_aware video multicast transport service over the Internet

Final Report Summary - P2P-PROVIDEO (P2P middleware for the deployment of an innovative business model for the provision of a QoS_aware video multicast transport service over the Internet)

During the first period of the P2P-PROVIDEO project a number of tasks have been accomplished. First, fine granularity scalability encoding (FGS) has been successfully applied for video transmission. An MPEG-4 FGS encoder has been developed. The encoder has been implemented in Visual C+ + using the Intel Integrated Performance Primitives (Intel IPP). Thus, a live video broadcast platform where a video source distributes a video stream to a number of clients in a multipoint fashion has been proposed. Multipoint communication has been achieved by applying a P2P approach, configuring a tree-structured overlay network where the root is the video source, while the other clients are internal nodes or leaves. We have also considered fine granular scalability (FGS) video streaming over a wireless ad hoc network composed of N nodes where each node is able to forward packets to the others. Video packets have been transmitted using Network Coding and efficiency of this scheme for this platform has been proved. Another video encoding taken under study has been the multiple description coding (MDC). We have successfully modified our live streaming system in order to transmit video packets using the multiple description coding. We have thus implemented together with Professor Lorenzo Favalli of the University of Pavia the ILPS-MDSC algorithm using the inter-layer prediction tools of the standard H. 264/SVC coder. Our developed architectures have a tree-based overlay topology which privileges end-to-end requirements minimising delays from the source, and avoids complexities like multiple tree management, coding inefficiencies due to multiple description coding, and problems ofout-of-order packets due to mesh structures. We have defined a Markov-based analytical model to evaluate performance and provide the designer with a tool to choose some project parameters. We have also explored the problem of the business model needed to support the economical growth of providers (video source), which would like to provide this kind of service. Thus, we have considered the users divided into two classes, according to the way they intend to pay the service: with bandwidth (cheap-tariff peers), or with some money (full-tariff peers). A revenue model for providers of this kind of services has been proposed to allow them to design their networks in order to maximise their revenue while satisfying user requirements in terms of both admission rejection probability and perceived quality on the received video stream.

During the second period of the P2P-PROVIDEO project a number of tasks have been accomplished. First, fine granularity scalability encoding (FGS) has been tuned in order to work within a P2P network. The final system [3, 8] has been tested on a small network made of up to ten PCs. The P2P approach has been set up configuring a tree-structured overlay network where the root is the video source, while the other clients are internal nodes or leaves. Another video encoding we have analysed has been the Scalable Video Encoding (SVC). We have developed a P2P architecture within the Linux platform using ffmpeg libraries and using the same P2P approach of above. Again, the system has been tested over a local network made of up to 10 PCs.

We have also defined a platform for multipoint real-time video distribution in the current Internet using multiple description encoding [5, 9, 13]. In particular, we have applied scalable multiple description coding, hierarchical video streaming and topology management as a solution to tackle problems of bandwidth variation and end-to-end delay minimisation.

Moreover, a new mechanism called Active Window Management (AWM) has been implemented in the access gateways [10]. This mechanism belongs to the second class of protocols because it does not require any modification either in the TCP protocol or in host implementations. It has been defined to maintain the queue length very close to a given target value and, at the same time, avoid packet losses in the gateways. By manipulating the Advertised Window parameter in the TCP ACKs, AWM gateways minimise the delay jitter and maximise the goodput of TCP sources. The AWM gateway action has the purpose of making the Advertised Window parameter more constraining than both the Congestion Window and the receiver Advertised Window each time the queue length exceeds the target value, thus driving the TCP sending rate. Let us stress that the AWM never overwrites the Advertised Window field with a value greater than the original one in order to not interfere with the original flow control algorithm of TCP. Moreover, the TCP sender works as usual; that is, it determines the sending rate by taking whichever is the lower between the Advertised Window and the Congestion Window. Let us note that the algorithm is very simple, and the processing power necessary to implement the AWM algorithm is similar to the processing power of any existing AQM technique, so it does not affect router performance appreciably. In addition, no per-flow state storage is needed in AWM gateways, ensuring scalability as number of flows increases.

It has been important to test in real environment the systems we have developed. Thus we have bought some NetFPGA platform (programmable routers) from Accent Technology. The NetFPGA platform allows for the rapid prototype and development of multi-Gigabit/second line rate networking applications. Thus, first, we have first analyzed the NetFPGA environment, then we have been able to start playing with it and send our coded video in a local area network at the University of Catania composed of two NetFPGA routers and several connected PCs. In the same network we have connected two Plug Computers (which we bought within the project) in order to have several heterogeneous network sources. We have also produced several papers within the NetFPGA platform [2, 4, 6, 10, 11, 12, 14] acquiring a lot of knowledge on it.

As far as the business model is concerned, we have also explored such a problem in order to support the economical growth of providers (video source), which would like to provide this kind of service. Thus, we have considered the users divided into two classes [1, 7], according to the way they intend to pay the service: with bandwidth (cheap-tariff peers), or with some money (full-tariff peers). A revenue model for providers of this kind of services has been proposed [1, 7] and developed to allow them to design their networks in order to maximise their revenue while satisfying user requirements in terms of both admission rejection probability and perceived quality on the received video stream. Dr. Reforgiato has been able to organise a team of people made of researchers and students: activities of knowledge integration and transfer have been organised weekly. In particular, two of his students (Enrico Santagati and Dr. Carla Panarello) have been hired to help out with the implementation of the architectures. Dr. Reforgiato has also been involved within an European project NAPA-WINE for the design of a new generation of P2P-HQTV systems and within an European project called ECONET for the design of router with low energy consumption. These activities further helped Dr. Reforgiato training activities/transfer of knowledge activities/integration activities. Two more NetFPGA cubes have been purchased from DBA Accent Technology in order to try the developed video encoding and overlay network techniques for scenarios like video streaming.

Dr. Reforgiato has been able to organise a team of people made of researchers and students: activities of knowledge integration and transfer have been organised weekly. In particular, three of his students have been hired to help out with the implementation of the live architectures (Sergio Barbera, Enrico Santagati and Dr. Carla Panarello). Dr. Reforgiato has also been involved within an Italian project (PRIN) called SORPASSO: works about the overlay network topology and management protocols have been done and comparisons between trees and meshes approaches have been performed. Within the SORPASSO project Dr. Reforgiato was able to get in touch with Professor Raffaele Bolla of the University of Genova, Professor Davide Adami of the University of Pisa and Professor Andrea Detti of the University of Rome.

Dr. Reforgiato has also been involved within the WP4 of the European project NAPA-WINE for the design of a new generation of P2P-HQTV systems. These activities further help Dr. Reforgiato training activities/transfer of knowledge activities/integration activities. Dr. Reforgiato was able to take contacts with several academic and industrial partners participating to this project: in particular with Professor Emilio Leonardi and Professor Marco Mellia of the University of Turin and Dr. Anastasios Zafeiropoulos of GRNET.

Moreover, a total of three NetFPGA cubes have been purchased from DBA Accent Technology in order to try the developed video encoding and overlay network techniques for scenarios like video streaming and in real environment.

As far as the use of the resources are concerned, there has been a small deviation with respect to the planned allocation. In particular, travel costs and publication costs have been lowered in favor of the other cost (personnel costs) and overheads. This has been done in order to hire three brilliant students, Sergio Barbera, Enrico Santagati and Carla Panrello who have developed part of proposed live P2P systems.