Since in some architectures the source still distinguish itself for reliability and resources made available to the service, some prefer using the term cooperative rather than Peer-to-Peer, in the sense that the users are not considered peers, but clients that cooperate in order to alleviate the load of the sever. Once it has received a chunk, a user can trade it with anyone else interested in it against another chunk. In this kind of application, the source splits the content into small blocks, termed chunks, that are transmitted to each end-user in parallel. Intra-session network coding, implemented at the application layer with little effort, has been proven beneficial for large-scale Peer-to-Peer (P2P) content distribution. The latter usually allows combination of packets from different sessions (multicast or unicast), thus referred to as inter-session network coding. Usually, the former approach is followed when coding is allowed only among packets belonging to the same multicast session, referred to as intra-session network coding.
Most of the solutions available perform network coding either at the Application Layer (OSI Layer 7), or at the Data-Link Layer (OSI Layer 2). Little emphasis has been given on the several options that the ISO/OSI or the TCP/IP protocol stacks offer regarding the level wherein NC should be integrated. So far, we have discussed the topic of network coding from a graph-theoretical point of view. On the other hand, a large size of the generation affects positively the performance of the network coding in terms of throughput, as more symbols are coded together (the Max-Flow Min-Cut Theorem assures that the maximum multicast flow is achieved only for h → ∞). Furthermore, the generation size also determines, together with the field size, the overhead due to the inclusion of coding vectors in the packet headers, which should be kept small, and particularly so in wireless networks, where packets are smaller and the overhead may become prohibitive. On one hand, the decoding delay is negatively affected by a large generation size k, since a sink node has to wait for k innovative vectors in order to decode. The choice of the generation size is not trivial, as it involves a trade-off between decoding delay and rate overhead, vs. The operations needed to recover the source message can be found at the receiver by Gaussian elimination, as soon as k innovative ( i.e., linearly independent) coding vectors have been received. When a sending opportunity occurs, a new packet is generated by combining, with random coefficients, all the packets in the current generation, and adding the coding coefficients in the packet’s header. In those nodes, the received packets are stored in separate buffers per generation. Only packets coming from the same generation can be combined at intermediate nodes. In PNC, if the source message is composed of many units, as it is often the case in a real scenario, in order to reduce the number of coefficients in each coding vector, the data stream is divided into generations, each one consisting of k consecutive data packets of fixed size, with k much smaller than the size of the message. Pesquet-Popescu, in Academic Press Library in Signal Processing, 2014 5.11.8.2 Practical Network Coding
0 kommentar(er)
