28-05-2013, 12:53 PM
A hybrid mesh-tree peer-to-peer overlay structure for layered video streaming
A hybrid mesh-tree.pdf (Size: 1.12 MB / Downloads: 56)
Abstract
Due to complicated implementation of IP multicast,
application layer multicast has attracted the attention of
researchers. Tree overlay has some challenges such as resiliency
problems caused by peer churns and also network resources may
not be efficiently used. Mesh overlay doesn’t have such problems,
but has a high delivery latency and overhead caused by pull
mechanism. Using a hybrid mesh/tree overlay offers push pull
approach and leads to use the advantages of both overlays
simultaneously. Also in p2p networks a significant problem is
heterogeneity of peers. In this paper we propose a hybrid
mesh/tree overlay that by means of streaming layered coded
video, addresses these challenges. This overlay poses other
critical design challenges such as latency and resiliency issues
that stable nodes in a distributed algorithm form a tree in order
to accelerate media streaming.
INTRODUCTION
Some drawbacks of traditional client-server networks
such as heavy overhead on sever have moved network
technology toward a distributed network model known as
P2P networks. In P2P networks peers have both roles of
client and server, so the server load is distributed between
peers.
Last studies on P2P file sharing shows that the majority
of traffic in internet is caused by files with an average of
1GB [1]. This means that most of requests are for video or
music files. Nowadays live streaming has become an
attractive approach in P2P networks for media delivery. The
most basic concept behind streaming is multicast, when the
source sends some chunks of data to some other nodes, but
multicast has two contradictions with multimedia streaming.
The first one is heterogeneity in peers download
bandwidth that leads to difference in their video quality
requirement, but multicast stream cannot meet everyone’s
requirement. The solution to this problem is layered coding
in which base layer provides basic quality for all peers and
for better quality peers request more enhancement layers.
RELATED WORK
The common topologies that can be used for media
streaming are tree, mesh or hybrid tree/mesh overlays. Some
of hybrid overlays that uses the characteristics of both pure
mesh and tree overlays that are proposed for live media
streaming are Anysee2 [2], HOWTO [3], mTreebone [4],
hybrid MCT [5] and MultiPeerCast [6].
Anysee2 uses a hybrid overlay consists of two overlays;
Control Tree which transmits the control message (messages
that contain peer selection and time synchronization
information) and Data Mesh in which media data is
transferred and tree in this overlay is constructed based on
peers’ locality. The main drawback of this mechanism is
that it doesn’t consider tree optimization, so the depth of
tree may get higher, therefore leaf nodes receive chunks of
data late and transmission delay would increase.
Mtreebone is another hybrid mesh/tree overlay that uses tree
for delivering media data and also control messages. And
mesh overlay here is used for transmitting missing data. In
spite of Anysee2, mtreebone considers a tree optimization
algorithm, but there are still some shortcomings. The main
drawback of mtreebone is that this overlay doesn’t maintain
the balance in tree and peers with large bandwidth may
crowd at some geographical areas.
PROPOSED METHOD
The most significant goal in p2p streaming are efficiently
using upload bandwidth of peers and transmitting media
with highest quality and throughput. For using the upload
bandwidths of peers and other network resources efficiently,
we have used a mesh overlay. In tree overlay, leaf nodes
cannot use their upload bandwidth. In our method we
offered a 2-level hybrid mesh/tree overlay, that by
transmitting layered coded video over this overlay
maximizes the quality and can address heterogeneity
problem of multimedia multicast.
The basic overlay here is mesh overlay. After a period of
time, the tree overlay will e formed for faster transmission
of layered coded video.
Mesh Construction
Every node at first joins to the mesh overlay in order to pull
the base layer and also enhancement layers of video through
this overlay.
The node willing to join the network, sends a message to
the tracker (a peer management server) that contains its
upload bandwidth. Based on this information the tracker
searches for peers with available bandwidth and makes a
response message that contains their address to the
requesting peer. Requesting peer establishes connection to
these selected nodes and receives chunks of data after
exchanging their buffer maps.
Tree Construction
When a node joins to the mesh overlay, it doesn’t have
the permission of joining to tree unless after of a period of
time becoming stable.
When the lifetime of this newly joined node gets equal or
more than a threshold time, this means that this node is a
stable peer and can become a tree member. Each stable peer
sends a join request to tracker in order to become a tree
member and then the tracker searches for suitable parent.
PERFORMANCE EVALUATION
The main goal of streaming protocols is to increase
quality and also to decrease end to end delay, so we focus
on these parameters as the evaluation parameters.
The simulation tool in this project was omnetpp-3.3p1. In
the simulation time of 100s, the source encodes and sends
about 3024 frames of an Fine Granular Scalable (FGS)
coded trace file video [10] having one base layer and four
enhancement layers. The encoding rate was 30 fps and
session time was about 100.8 seconds. The bandwidth of
peer has been selected in a manner that half of them can
receive four enhancement layers and half of them can
receive just two of them. The average rate of base layer and
enhancement layer was adjusted to 2 Mbps and 200 kbps.
We had compared our method with a mesh overlay
because our method is an improvement of mesh topology.
The mesh overlay that is compared to our method has
exactly the characteristics of our hybrid method. This means
that this overlay is exactly the mesh overlay that is
constructed in the first phase of our topology construction
and the video transmitted over this overlay was the layered
coded media that was used in our hybrid overlay.
CONCLUSION
In this paper we have used the advantages of tree and
mesh overlay in order to prose a hybrid mesh/tree overlay.
For solving the most significant challenge of P2P
networks, peer churn, we selected stable peers in order to
form a tree and inject the base layer of a layered coded
media to the mesh overlay. For using network resources in
a more efficient way we have used mesh overlay.
In future we want to improve our method by considering
the peer locality in neighbor selection in tree and mesh
overlay in order to reduce loss and delay and compare the
designed method to the existing hybrid protocols. Also
using an incentive mechanism can be useful to reduce free
riding in P2P networks.