An overview of peer-assisted live streaming systems: In peer-assisted live streaming systems, the streaming server(s) stream data to some of the users. This ensures that an increase in the number of users also increases the system's streaming capacity.
Motivation for the Research Work
Poor resistance to peer duplication and improper use of peer load capacity in tree overlays led to the development of multi-tree and mesh overlays. In the existing literature, the stability of peers and the contribution of their load capacity are predicted using session duration [56].
![Figure 1.1: Surge in number of users in Skype in 2010 [1]](https://thumb-ap.123doks.com/thumbv2/azpdfnet/10551675.0/28.892.124.740.153.811/figure-1-1-surge-number-users-skype-2010.webp)
Contributions of the Thesis
A Peer Selection Strategy for Mesh Overlays
During peer selection, we consider parameters such as upload capacity, propagation delay, chunk buffer duration and buffer level. In the second phase (peer tier), a peer selects its partners from those suggested by the tracker, taking into account their upload capacity, propagation delay, chunk buffer duration and buffer level.
An Overlay Management Strategy for Hybrid Tree- Mesh OverlaysMesh Overlays
The streaming quality experienced by peers is also improved by 25% because seeders with higher upload capacity are created as part of the expanded CDN tree, and partners with high stability, chunk availability and free upload capacity are chosen. Peer upload capacity utilization is increased by 30% using the proposed strategy because it strategically uses the upload capacity of all peers with heterogeneous session duration and upload bandwidth.
Organization of the Thesis
The stability of peers is improved by 20% because they get better stream quality and lower startup delay. Chapter 5 presents an overlay management strategy for the hybrid tree mesh overlay to improve the upload capacity utilization of peers and to improve the QoS despite the heterogeneity of peers.
Architecture of a Peer-assisted Live Stream- ing Systeming System
Some of the popular peer-assisted systems using mesh overlay are PPlive [21], Coolstreaming [22] and Fast-Mesh [51]. Hybrid Tree Mesh Overlays: In hybrid tree mesh overlays, some of the peers are a part of the tree topology while the rest are connected to each other using a mesh topology, as shown in Fig.
Literature Survey
- Impact of Overlay Topologies on QoS
- Peer Selection Strategies
- Chunk Scheduling Mechanisms
- Organizing Peers Considering Heterogeneity
In [52], it was proposed to keep peers with higher upload capacity closer to the source to improve the perceived stream quality and playback delay. In the literature, most of the existing systems organize peers based on their stability and upload capacity.
Summary
Introduction
None of the existing works compared multi-tree, mesh, and hybrid tree mesh overlays in terms of transmission quality, playback delay, start-up delay, and stability during streaming. It also provides insights into the organization of peers in the tree and mesh hybrid overlay tree mesh taking into account their heterogeneity in lifetime and bandwidth. Hybrid tree mesh overlay results in slightly lower start-up delay, playback delay, and transmission quality compared to mesh overlay.
Construction and Management of Overlay Topologies: An OverviewTopologies: An Overview
Multi-tree Overlays
To join multiple trees, new peers find a parent in each tree using server or root of the tree. In CoopNet [63] the server finds a parent for the new peer in each tree, while in [64] new peers randomly choose a parent to download a portion of the stream. To avoid creating a cycle, each peer of the spare capacity group maintains the information of path from the root.
Mesh Overlays
Each peer selects its partners to receive broadcast parts from the list of active peers. Each peer first creates a plan to request chunks with the help of its partners' buffer maps. However, peers in most mesh overlays identify the sudden departure of partners by monitoring the share loading rate or periodic messages [91, 92].
Hybrid Tree-Mesh Overlays
In [66], initially all new peers are added to the mesh topology at the top layer up to a specific hopcount threshold. Data delivery mechanism: Peers joining tree topology (tree peers) use push-based mechanism, while mesh peers use either pull- or push-pull-based data delivery mechanism. In [96], peers use push- or pull-based data delivery depending on the duration of a connection.
Performance Evaluation
Evaluation Scenarios
The upload capacity distribution of peers is given in table 3.3, where 30 percent of peers are free riders. To simulate peer attrition, the lifetime or session duration of peers is set using Pareto distribution [56, 102]. It represents that peer age is an indicator of their expected remaining time in the system.
Evaluation Metrics
Stream recovery duration: This is the length of time for which peers experience degradation in stream quality before parent/partners are re-elected during peer departure and the desired quality is regained. Playback delay: This is the time it takes to deliver chunks to a peer from the source. Chunk buffering duration: This is the length of time a chunk is buffered at a peer before being forwarded to another peer.
Results and Discussion
On the other hand, the higher buffering delay in the mesh overlay increases the startup delay compared to the hybrid overlay. We can see that a higher proportion of peers experience a clustering error in a multi-tree overlay. It is observed that peers in mesh and hybrid overlay can recover their streaming quality faster during failover compared to multi-tree overlay.
Summary
Introduction
In this chapter, we propose a three-stage peer selection strategy to reduce the replication delay and startup delay experienced by peers in mesh overlays. In the first phase (tracker-nier), the tracker selects potential partners taking into account the load capacity and buffering level of peers. We compare the performance of the proposed strategy with Fast-Mesh [51] and HLPSP [60], both with and without bottleneck.
Proposed Peer Selection Strategy
System Model
In the second stage (peer), a peer selects its partners from those suggested by the tracker based on upload capacity, propagation delay, chunk buffer duration and buffer level of the partners. When a new peer joins the system, the tracker suggests few prospective partners to the new peer using a peer selection strategy (tracker level). After receiving a list of prospective partners from the tracker, a new peer selects its initial partners using a different peer selection strategy (peer level).
Proposed Three-stage Peer Selection Strategy
If the total load capacity of the new peer is greater than zero, then the tracker adds the new peer to the LPL. In the second part, the delay of the last hop, i.e., the path from the potential partner to the new peer, is estimated. The last hop transmission delay is estimated with the help of the remaining load capacity of the potential partner.
Computational Complexity
When an ancestor receives a request message, it compares its total upload capacity with the peer's total upload capacity. If the total upload capacity of the ancestor is less than that of the peer, the ancestor responds with a response message. After receiving the response message from the ancestor(s), the peer calculates the rank of each ancestor using Eq.
Performance Evaluation
Evaluation Scenarios
We varied α from 0.2 to 0.8 and observed its impact on the average playback delay and startup delay by the peers, as shown in Fig. It is found that playback delay is minimum when α = 0.7, while startup delay is minimum when α = 0.5. This represents that the peers' age is an indicator of their expected remaining time in the system.
Evaluation Metrics
This is because the authors of [56, 102] have shown that peer lifetimes follow heavy-tailed distributions such as the Pareto and Weibull distributions. Peers arrive according to a Poisson process, where the time between peer arrivals is modeled using an exponential distribution, also used in [104–106] .
Results and Discussions
It is observed that the proposed strategy shows lower playback delay than Fast-Mesh and HLPSP in all conditions. When Fast-Mesh and our strategy use topology adaptation, the playback delay with the proposed strategy is reduced by 30-32%. This is due to significant improvement in buffer delay experienced by the peers in the proposed strategy.
Summary
The proposed strategy significantly improves the playback delay, while the streaming quality and start-up delay also need to be improved because they affect the lifetime and upload contribution of peers in the system. In the next section, we propose an overlay management strategy that organizes peers in the hybrid tree-mesh overlay to improve the streaming quality, startup delay, and upload capacity of peers while considering the heterogeneity in bandwidth and lifetime of peers.
Introduction
They also neglect to utilize the load capacity of peers with short session duration and/or low bandwidth. They also do not take into account the service of colleagues during their arrangement in the overlap, which depends not only on their durability, but also on the availability of parts, loading and unloading capacities as a whole. In this chapter, we propose a congestion management strategy that focuses on increasing peer load capacity utilization and improving QoS in terms of transmission quality and start-up delay.
Proposed Overlay Management Strategy
- System Model
- CDN Server Selection Strategy
- Creating and Maintaining Hybrid Tree-Mesh Overlay
- Creating and Maintaining Extended CDN Tree
- Creating and Maintaining Mesh Topology
- Overhead and message passing complexity
UAr (or UBr)← Residual upload capacity of server A (or B) Dna ← Cumulative download capacity of the new peer '. The number of peers in a virtual resource is equal to the number of substreams that the new peer can download. As explained earlier (see Sect. 5.2.4), the remaining upload capacity of a tree peer is reserved and used to create virtual resources until it becomes stable.
Performance Evaluation
Evaluation Scenarios
The CDN server then adds the new peer as a tree peer or a mesh peer in the overlay, after comparing the new peer's upload capacity to the server's directly connected child peers. Here, the extent of the CDN servers (Zc), the number of CDN servers (Z), and the number of potential partners (K) of a peer are negligible compared to the number of peers (N). The distribution of peer upload capacity is shown in Table 5.2, where approximately thirty percent of the peers are free riders and the total upload contribution from peers is approximately forty percent of the system upload capacity [14].
Evaluation Metrics
Peer stability: This is the ratio of actual session duration to the maximum session duration of the peer. The maximum session duration is the length of time between the peer join time and the end of the streaming session. Early exit rate: This is the percentage of peers that exit before the end of the session due to unacceptable startup delay and/or stream quality.
Results and Discussions
The startup delay increases with the number of peers due to the increase in power delivery load of CDN servers and existing peers. It is found that the proposed strategy improves the perceived streaming quality by efficiently utilizing the upload capabilities of peers. Therefore, both the stability of peers and the upload contribution of peers are improved using the proposed strategy.
Summary
It is clear that the upload contribution from peers is higher as well as closer to the ideal contribution. Better stability of peers also increased the upload contribution of peers to the system by about 15%. Peer upload capacity utilization is improved by 30%, which also reduced the stream delivery load of CDN servers by approx. 20%.
Conclusions
Finally, we addressed the problem of improving streaming quality, startup latency, and upload capacity utilization considering heterogeneous peer lifetimes and bandwidths. We compared the performance of the proposed strategy with existing systems and found that it significantly improved the streaming quality and startup delay of peer systems. The proposed strategy also significantly increased the utilization of peer-to-peer upload capacity, leading to offloading of CDN servers.
Future Work
Khonsari, "Playback Continuity Exploration and Delay Exchange in Peer-to-Peer Transmission," in Proc. Rejaie, “PRIME: Peer-to-peer Receiver-driven-Mesh-based Streaming,” IEEE/ACM Transactions on Networking, vol. Yum, “CoolStreaming/DONet: A data-driven overlay network for peer-to-peer live media streaming,” in Proc.
