Image Processing Reference
In-Depth Information
(e.g. reduce delay, jitter, etc.), (b) reduce the control overheads, (c) significantly
increase the performance of the overall network and reduce the network bandwidth
requirements and (d) reduce the network infrastructure operational costs.
In the literature, there are many overlay construction algorithms that try to
optimize various criteria such as to minimize the average cost of shortest path
trees rooted at group members, minimize the overlay topology diameter while
respecting the degree constraints, minimize the root-path latency for a specific
node, minimize the average delay and maximize the average bandwidth. We
summarize and propose the following high level algorithm:
• Step 1: Network nodes, which are members of the overlay will keep a connected
mesh topology among themselves following one of the many approaches avail-
able in the literature [
3
]. The mesh topology is explicitly created at the beginning
of a session, based on various “cost” metrics.
• Step 2: Whenever a new client joins a session, a spanning tree will be created
over the mesh topology and will be optimized based on a number of criteria
(mainly avoidance of loops, while keeping the minimal latency and the average
bandwidth requirements).
• Step 3: Depending on the assumed life-time of the sessions, a
Spanning Tree
refinement
procedure will periodically run to improve the overlay performance
as well as to deal with fluctuations in the available network resources and
congestion. Yet, as the effectiveness of the refinement to real-time applications
is questionable, due to interrupted data distributions among the members, this
approach will be further studied at deployment time.
In all three steps, the “cost” of an overlay path between any pair of nodes should
be comparable to the “cost” of the unicast path between that pair of members.
12.3 Constructing the Network Overlay
In order to construct an overlay topology various direct and implicit (network
related, node related, network operator related, social network related) metrics
should be considered to select the “best” nodes for each client per session or the
“best” parent nodes, during the construction of a multicast overlay. As “best” node,
we define the node that provides the optimal average PQoE to the 3D immersive
communications clients that participate in the session, with minimal network
overhead and operational cost. In parallel, the proposed solution should avoid
approaches that drastically increase the network load (e.g. such as P2P techniques)
and decrease the overall efficiency of the network. Last but not least, we should take
into account, metrics related to the cost of the network, if applicable. As a result, a
number of parameters should be taken into account, such as:
(a)
Network Characteristics.
We assume a number of metrics such as the end-to-
end available bandwidth,
the end-to-end network delay, one-way delay,
