Study Of Bandwidth Utilization Optimization For P2P Live Streaming Systems

In P2P systems, peers help each other to leverage user resources. As for live streamingsystems, network bandwidth resource is the key scarce resource. So, how to maximize theutilization of the bandwidth resources and how to minimize the inter-AS traffic are the keyissues in the design of the P2P streaming systems.To maximize the bandwidth we need to refine the data scheduling algorithm. A datascheduling algorithm decides for a peer which data segments to fetch and where to getthem. So it affects greatly the performance of a peer. Peers in P2P live streaming systemsare categorized into two kinds: one is ordinary peer, which consumes streaming data forplayback, the other is volunteer peer, which just helps others getting their data and doesnot play the streaming data. To maximize the bandwidth utilization, two different datascheduling algorithms should be designed according to the two kinds of peers.ColorStream is proposed for the ordinary peers. When deciding which segments to get,in addition to the rarity and urgency considered in existing data algorithms, ColorStreamalso considers the freshness of segments due to fresh segments having longer time forsharing. Those segments stay in the sliding window a longer time are assigned with greaterpriorities. Furthermore, ColorStream can bring more benefits to peers when the networkbandwidth between peers are low and the number of connections used are limited.ContriMax, a heuristics data scheduling algorithm is also designed for volunteersto collect and maximize the contributions of them. Some networks contain thousandsof semi-rider users (ADSL-users), those peers with very limited outbound bandwidth,contribute much less than what they obtain in terms of data. This kind of environment putshigh stress to the data source even crashing the whole system. But in the VolunteerLivearchitecture, those idle peers are encouraged to contribute their network bandwidth thoughthey do not enjoy the service. In VolunteerLive, volunteers employ ContriMax datascheduling algorithm to maximize their contributions.There are two methods to save inter-network traffic: optimize overlay structure andoptimize the data scheduling algorithm. The data travels a shorter path if the logical overlay matches the physical underlying network, which means decreasing inter-network traffic.Further, CoFetch, which is based on optimizing the data scheduling algorithm, is alsoproposed to decrease the inter-network traffic.In tree-based overlay, peers have low delay between each other. However some peersusually have not enough bandwidth to support their children and it is difficult to resistthe churn. In mesh-based overlay, peers can resist the churn, but it is difficult to optimizethe network structure leading to redundant traffic. Tree-Control-Media-Mesh (TCMM)leverages the advantages of the both kind of overlay to decrease the inter-network traffic. InTCMM, transmissions of media data are controlled by two independent relay protocols ina collaborative manner. One protocol is used to help a peer to identify its neighbor peersusing the location information while the other one is used to deliver media stream amongthe peers. The two protocols organize all peers to two graphs with different topologies thatthe communications can benefit a lot from the hybrid control topology.In existing fetching data algorithms in P2P streaming systems, the peers usuallyfetch data independently and non-collaboratively which leads to huge inter-network traffic.CoFetch, a collaboratively data scheduling algorithm, decreases the inter-network trafficby letting the peers collaboratively sharing data locally and decreasing the data request tointer-network peers. In CoFetch, to avoid all peers fetching same data from inter-networkconnections and to make the best of the local bandwidth, instead of aggressively fetchingdata absolutely depending on their own desires, peers divide requests into multiple steps.By this way, peers can share greatly the data fetched in previous steps.To date, there are lots of academic and commercial live media streaming systems,which usually provide same programs to public. All the users are distributed randomlythroughout different networks and cannot share data across networks. Thus it leads toredundant inter-network (e.g., AS or ISP) traffic. PPUnion lets peers from differentoverlays, owning by different commercial service provider, search and share data. Both theinter-network traffic and the stress on the data source can be suppressed. At the same time,the performance of the peers is improved.