Sei sulla pagina 1di 8

(IJCNS) International Journal of Computer and Network Security, 79

Vol. 2, No. 9, September 2010

Performance Evaluation of Peer–to-Peer system in


Video on Demand with Load Sharing Scheduling
Algorithm
S.Nagaprasad1*, A.VinayaBabu2, K.Madhukar3, D.Marlene G Verghese4, V.Mallaiah1and A.Sreelatha5
1
Research Scholar, Dept. of CSE,
Aacharya Nagarjuna University, Guntur,
1
*nagkanna80@gmail.com
2
Professor, Dept. of CSE,
JNTU College of Engineering, Hyderabad,
avb1222@gmail.com
3
Professor, Nizam College, Osmania University, Hyderabad
mkatakam@osmania.ac.in
4
Associate Professor
Bhimavaram Institute of Engineering & Technology
Pennada, West Godavari
marlenedegala@yahoo.co.in
5
Student, Dept. of CSE,
JITS College of Engineering, Karimnagar,
sreelathu_alugoju@yahoo.com

Abstract: Peer-to-Peer (P2P) networks have been shown to be Currently, the traffic generated by P2P systems accounts
a promising approach to provide large-scale Video on Demand for a major fraction of the Internet traffic today, and is
(VoD) services over Internet for its potential high scalability. bound to increase. The increasingly large volume of P2P
However, for a normal peer, how to efficiently schedule media traffic highlights the importance of caching such traffic to
data to multiple asynchronous peers for VoD services in such reduce the cost incurred by Internet Services Providers
networks remains a major challenge. These systems
(ISPs) and alleviate the load on the Internet backbone.
dramatically reduce the server loading, and provide a platform
We are faced with a problem of delivering quality video
for scalable content distribution, as long as there is interest for
to a single receiver computer. In the streaming scenarios,
the content. The main challenges reside in ensuring that users
can start watching a movie at any point in time, with small start- the entire video is not always available at every peer
up times and sustainable playback rates. In this work, we machine and/or it would not be feasible to transmit the
address the challenges underlying the problem of near Video- entire video from a single peer ,for example, that would
on-Demand (nVoD) using P2P systems, and provide evidence overload a particular peer. The beauty of streaming is
that high-quality nVoD is feasible. In particular, we investigate obviously the fact that we don’t need to have the entire video
the scheduling problem of efficiently disseminating the blocks of downloaded before the play out begins [6][10]. We can
a video file in a P2P mesh-based system, and show scheduling simply split the video file, identify the peers that have the
algorithm can provide significant benefits and the experimental segments of interest available, request these segments from
results will show that load sharing scheduling performs
the peers, receive them and play them out.
significantly better than other dynamic algorithm, network
An important requirement of a VoD service is scalability,
coding.
i.e., to be able to support a large number of users, as a
Keywords: Networks, VoD, nVoD, P2P. typical video stream imposes a heavy burden both on the
1. Introduction network and the system resources e.g. disk I/O of the server.
The multicasting paradigm has been proposed to address the
Video-on-demand (VoD) systems provide multimedia scalability issues. However, these systems require a
services offering more flexibility and convenience to users multicast-enabled infrastructure, which unfortunately has
by allowing them to watch any kind of video at any point in never materialized. Peer-to-Peer networks promise to
time. Such systems are capable of delivering the requested provide scalable distribution solutions without infrastructure
information and responsible for providing continuous support.
multimedia visualization [5].
80 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 9, September 2010

The difficulty lies in the fact that users need to receive (d)Other servers such as log servers for logging significant
blocks “sequentially” in order to watch the movie while events for data measurement, and transit servers for helping
downloading, and, unlike streaming systems, the users may peers in the system [11].
be interested in different parts of the movie, and may We assume a large number of users interested in some
compete for system resources. In other words, we assume video content, which initially exists on a special peer that
linear viewing, but we allow the users to join at arbitrary we call the server. Users arrive at random points in time,
times. The resources especially network bandwidth of the and want to watch the video sequentially from the beginning
server are limited, and hence users contribute their own of the server are limited, and hence, users should contribute
resources to the system. Users organize themselves in an their own resources to the system. The upload and download
unstructured overlay mesh which resembles a random capacities of the users are also limited and typically
graph. The goal then is to design a P2P system which meets asymmetric. A client joins the system by contacting a central
the VoD requirements, while maintaining a high utilization tracker. This tracker gives the client a small subset of active
of the system resources [10]. nodes .The client then contacts each of these nodes and joins
We study algorithms that provide the users with a high- the network. At any point in time, a node is connected to a
quality VoD service while ensuring a high utilization of the small subset of the active nodes, and can exchange content
system resources. We evaluate our algorithm such as and control messages only with them. We call this subset
segment scheduling using both extensive simulations and the neighborhood of the node.The neighborhood changes as
real experiments under different user arrival/departure a result of node arrivals and departures, and because nodes
patterns. The results will show that load sharing scheduling periodically try to find new neighbors to increase their
algorithm will be able to improve throughput for bulk data download rates. We assume cooperative nodes. The file is
transfer and scheduling that results in high system divided into a number of segments, which are further
throughput while delivering content “pseudo sequentially”, divided into blocks. The system is media codec agnostic,
to provide efficient VoD with small setup delays as hence, nodes need to download all blocks; if a block is not
compared to other dynamic algorithms such as Segment available when needed, the video pauses and this is
Scheduling and Network Coding[1][2][6]. undesirable. Clients have enough storage to keep all the
In this paper we are implementing Scheduling which blocks they have downloaded.
takes Segment Scheduling in which the entire video is Our system divides the constant stream of data into
divided into segments once the play out begins. It takes stripes to improve performance and robustness. In a peer-to-
video in the form of segments, define the segments, identify peer system, the stream of data is dis-rupted whenever a
the peers that have segments of interest available, request client leaves the system either due to a failure or a regular
these segments from the peers , receive them and play them disconnects. Since clients receive pieces of the content from
out. Load sharing Scheduling algorithm will improve different senders, they can continue to receive some data
performance by increasing sender’s bandwidth, dividing even if one of the senders disconnects. To further hide
segments based on loads and the number of segments and disruption from the user, a client keeps a buffer of several
decreasing number of missed segments. Section 2 describes seconds of data. When a client reconnects after being cut off
Architecture of the peer to peer system with load sharing from a sender, the buffer allows the video to play smoothly
scheduling algorithm. Section 3 describes brief description as the client catches up on the data that was missed during
of algorithm. Section 4 describes the implementation of the the disconnection. In a one-directional live video streaming
whole algorithm, Section 5 describes simulations and system, it is allowable for the video to be viewed a few
Section 6 presents our conclusion. seconds after its creation.

2. Architecture and Environment


P2P-VoD system has the following major components:
(a)A set of servers as the source of content (e.g., movies);
(b)A set of trackers to help peers connect to other peers to
share the same content;
(c) A bootstrap server to help peers to find a suitable tracker
(e.g. based on which geographical region the peer is
located), and to perform other bootstrapping functions;

Figure 1. System Architecture of Video on demand


(IJCNS) International Journal of Computer and Network Security, 81
Vol. 2, No. 9, September 2010

The goals of our system include the following: Our algorithm hinges on having a good estimate of how
• To ensure low set up time and a high sustainable well-represented a segment is. This estimate should include
playback rate for all users, regardless of their nodes that have the complete segment, and those that have
arrival time. partially downloaded the segment. In our implementation,
• We are also interested in increasing the total number the tracker monitors the rarity of segments in the network.
of blocks exchanged per round, which we call Clients in our system report the fraction of blocks they have
throughput, and the total server bandwidth. received from each segment. Those fractions are used to
estimate the popularity of the segments; for example, a
3. Related Work segment is considered under-represented if the vast majority
The overall throughput improves if all the nodes seek to of nodes have very few blocks from that segment [14] [15].
improve the diversity of segments in the network. If the The primary objective of the scheduling algorithm is to
segment policy is to upload a block from a lesser represented create a schedule such that all requested video segments are
segment whenever possible throughput improves delivered to the receiver before their respective deadlines. If
significantly for both existing and new nodes [15] [19]. that is impossible to achieve given insufficient resources,
We are faced with a problem of delivering quality video etc., we want to minimize the number of segments missed.
to a single receiver computer. In the streaming scenarios, We also have two secondary objectives. We want to make
the entire video is not always available at every peer efficient use of the bandwidth and avoid transmitting
machine and/or it would not be feasible to transmit the segments before they are required because there is no
entire video from a single peer. guarantee that they will even be used. Finally, since we do
It is important that we provide a continuous supply of the not want to overload any of the senders, we make sure that
video segments once the play out begins. If only a few the peers are load balanced. The sender for a particular
segments are missing once in a while, the video quality will segment is selected based on the estimated load of each
still be acceptable by most standards. If too many segments sender. The estimated load of a sender is the amount of
are missing the video and the audio quality will suffer and work time spent transmitting segments that the sender has
the media will not be useful. A missing segment does not assigned by the schedule at a particular instance of time.
always indicate network failure (due to congestion or The time it takes a particular sender to transmit a segment
other problems). It may simply mean that the segment was depends on the segment size and the sender bandwidth. We
not available at the receiver’s media player at the time it was predict the load of each sender by temporarily assigning the
supposed to be rendered, i.e. the segment deadline was next scheduled segment to each sender. We select the sender
missed. The main problem that we are solving is creating a with the smallest predicted estimated load. This ensures that
schedule that minimizes the number of video segments the load is shared among all the senders. The proper
missing their deadlines [14][15][18][19]. segment transmission start time is also most essential for a
Load Sharing Scheduling Algorithm provides track of successful schedule. We schedule each segment to arrive at
every sender’s estimated load. Before we can sort the list of the receiver a fixed amount of time before its deadline thus
suppliers for segments, where segment has n potential fixing the client buffer. This ensures the least amount of
suppliers, we create a look-ahead estimated load, i.e. what bandwidth wasted on unsolicited video segments.
would be the load of each one of those potential suppliers if In situations where senders do not have all segments
segment s was to be assigned to them. Once we have these available, we must make sure that we do not commit a
look-ahead estimated loads, we can sort the list of potential sender who is the only one that can deliver particular
suppliers in the increasing order of the estimated load. Since segments to do work delivering other segments that could be
we do not have a guarantee that the supplier with the delivered by other senders. When determining the
smallest estimated load is suitable to deliver segments, we appropriate peer to deliver a segment, we must therefore
iterate over the list of potential suppliers. Algorithm examine the number of potential senders that have the
minimizes the number of missed segments and the easiest segment available. It is usually more difficult to meet the
way to do this is to compare it against a couple of other deadline of a segment that has fewer potential suppliers [4]
known scheduling algorithms and the results show that that [10] [14] [15]. For this reason, the algorithm first calculates
load sharing scheduling improves the throughput of the the number of potential suppliers for each segment and the
system and decrease missed segments[20]. segments with least potential suppliers are scheduled first.
That is, the segments with one potential supplier are
4. Load Sharing Scheduling Algorithm scheduled first, then segments with two potential suppliers,
82 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 9, September 2010

then three and so on. This is a potential area of improving bandwidth decreases the number of missed segments which
the algorithm in the future because it is not always necessary solves the problem of missing segments. It’s possible that
to schedule the segments with the least potential suppliers we will have any number of segments with n potential
first. It is still possible to change segment scheduling order suppliers; we must iterate over the list of segments with n
from the order of the number of potential suppliers and still potential suppliers. If we have large-scale video, we can
create an efficient schedule where all segment deadlines are simply split the video file into well defined segments,
met. The proper segment transmission start time is also identify the peers that have the segments of interest
most essential for a successful schedule. We schedule each available, request these segments from the peers receive
segment to arrive at the receiver a fixed amount of time them and play them out. In situations where senders do not
before its deadline thus fixing the client buffer. This ensures have all segments available, we must make sure that we do
the least amount of bandwidth wasted on unsolicited video not commit a sender who is the only one that can deliver
segments [16] [18]. particular segments to do work delivering other segments
In the most general form we are given a set of n video that could be delivered by other senders.
segments, S = {S1, S2, . . . , Sn}, with an associated set of The first task is to create the segment supplier table. The
segment lengths, L = {L1, L2, . . . , Ln}, and a set of reason for creating this table is to enable actual schedule
segment deadlines, D = {D1, D2, . . . , Dn}. We also have a creation code to iterate the segments in the increasing order
set of m peers, P = {P1, P2, . . . , Pm} with an associated set of the number of segment suppliers. The Segment table
of peer bandwidths, B = {B1, B2, . . . , Bm}[20]. Finally, we should fetch data in such a way that it can be easy to
are given the set of segment ranges available at each peer. retrieve. For every potential sender p for segment s, we must
Since typically peers hold contiguous ranges of segments, confirm that the sender p is suitable to deliver segment s.
we assume that the segment range at each peer is simply This means that sender p must be able to deliver segment s
given by the highest segment number available. We assume before its deadline and it cannot overload the receiver, i.e.
that all segments with lower segment numbers are also the resulting bandwidth at the receiver must be ≤ maximum
available at that peer. The set of available segments is given Bandwidth. Secondly, we create schedule that minimizes the
by A = {A1, A2, . . . , Am}. We need to create a schedule J number of video segments missing their deadlines.
where each segment has an assigned peer and transmission Increasing the average sender bandwidth has very little
start time such that the segment will be transmitted to the effect on the execution time but a higher average sender
receiver before its deadline. Since this may not always be bandwidth decreases the number of missed segments which
possible and missed segments must be taken into solves the problem of missing segments. It’s possible that
consideration, we want to minimize the number of segments we will have any number of segments with n potential
that miss their deadlines. We have the additional constraint suppliers; we must iterate over the list of segments with n
that the schedule cannot exceed the incoming receiver potential suppliers. This data is readily available from the
bandwidth. segment supplier table. For every segment with n potential
Consider an instance of the segment scheduling problem suppliers, we need to determine its sender and transmission
where we are given a set of n segments. start time. Since we want to schedule the segments with
S = {S1, S2,..., Sn} and a set of m peers least potential suppliers first, we must iterate over the
P = {P1, P2,..., Pm}. We apply two restrictions: collection of segment supplier table indices in the increasing
1) There is only one sender peer available, i.e. m = 1. table index order. It’s possible that we will have any number
2) The sender peer has all the segments available. of segments with n potential suppliers; we must iterate over
The segment transmissions are the tasks that need to be the list of segments with n potential suppliers. This data is
scheduled. The release time of each segment transmission is readily available from the segment supplier table. For every
0. Since we only have one sender, clearly only one segment segment with n potential suppliers, we need to determine its
can be transmitted at a time. sender and transmission start time.
The execution length of each task is the segment Here, schedule creation iterates the segments in the
transmission time, which can be calculated from the increasing order of the number of segment suppliers. The
Segment length and the sender’s bandwidth. Every segment data is arranged in such a way that it is easy to retrieve. For
still has a deadline. At this point, it should be obvious that every potential sender p for segment s, we must confirm that
we have an instance of sequencing the Sender’s problem to the sender p is suitable to deliver segment s. Then, the
be solved by our Efficient Scheduling algorithm. segments are transferred from senders to the receivers with
Increasing the average sender bandwidth has very little the least segment first. The step 5-10 iterates the loop for
effect on the execution time but a higher average sender loading peers. The segments are then loaded at the client
(IJCNS) International Journal of Computer and Network Security, 83
Vol. 2, No. 9, September 2010

side with the increased bandwidth at the server side. The (ii) To pre-fetch or not to pre-fetch;
bandwidth of client should not exceed the maximum while pre-fetching could improve performance, it could also
available bandwidth at Server side. Finally, the scheduled waste uplink bandwidth resources of the peer.
peers are returned. (iii) Selecting which segment or movie to remove when the
For every potential sender p for segment s, we must disc cache is full; preferred choices for many caching
confirm that the sender p is suitable to deliver segment s. algorithms are least recently used (LRU) or least frequent
This means that sender p must be able to deliver segment s used (LFU).
before its deadline and it cannot overload the receiver, i.e. Content Discovery:
the resulting bandwidth at the receiver must be ≤ maximum Together with a good replication strategy, peers must also be
bandwidth. As the initial guess, we set the start time to be able to learn who is holding the content they need without
such that segment s will arrive at the receiver right before its introducing too much overhead into the system. P2P systems
deadline and before we confirm this transmission slot for depend on the following methods for content discovery:
segment s we must make sure that it doesn’t violate the (i) A tracker; to keep track of which peers are replicating
maximum bandwidth constraint. what part of the movie;
(ii) DHT; used to assign movies to trackers for load
balancing purposes.
Congestion Control
• Rate Control-Match the rate of the video stream to the
maximum available bandwidth thus reducing congestion
and segment loss. Without rate control segments which
would exceed the maximum bandwidth would be discarded.
This approach focuses on the transport concept.
• Rate-adaptive video encoding -Use compression to make
video feeds more practical and bandwidth efficient. This
approach focuses on the compression concept.
• Rate Shaping -This approach is a combination of the
previous two. The video feed is being re-coded with rate-
adaptive video encoding and rate control makes sure there is
no loss in segments.

5. Performance Evaluation
In this section we evaluate the performance of both
segment scheduling and load sharing scheduling algorithm.
Initially, we will present data that compares the segment
rate of both the algorithms. Then Load sharing Scheduling
algorithm will be compared with another dynamic
Figure 2. Load Sharing Scheduling Algorithm Replication algorithm. And then the results will clearly show that load
Strategy of Segments sharing scheduling algorithm is better than segment
scheduling especially under the missed segments.
Assuming each peer contributes with some amount of hard
5.1. Simulation Set Up:
disc storage, a P2P storage system is formed by the entire Matlab Software was used for all simulations. Along with
viewer population, each of them containing segments. Once load sharing scheduling algorithm network coding and
all the segments are available locally, the segment is segment scheduling will also be simulated for benchmark
advertised to other Peers. The aim of replication strategy is comparison purposes. In general, the behavior of the
to make segments available to every user in the shortest time algorithms is rather intuitive. For all 3 algorithms,
possible in order to meet with viewing demands. Design increasing the number of senders has very little effect on
issues regarding replication strategies contemplate: the execution time but having more senders’ decreases the
(i) Allowing multiple movies to be cached if there is room number of missed segments. Similarly, increasing the
on the hard disc. This is referred as multiple movie cache average sender bandwidth has very little effect on the
(MVC); and lets a peer watching a movie upload a different execution time but a higher average sender bandwidth
movie at the same time.
84 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 9, September 2010

decreases the number of missed segments. Finally, simulated scheduling algorithm with increased bandwidth
increasing the bit rate has similar effects, i.e. it has little on server side systems used in our experiments.
effect on the execution time but increasing the bit rate
increases the number of missed segments[11][13]. Table 1: System Parameters.
The parameters that we should vary include:
Parameters. Values
• Average segment size, i.e. video quality (bit rate) Segments 10
• Segment count (shorter vs. longer videos) Potential Suppliers 4
• Number of senders Bandwidth 100kbps
• Bandwidth of senders bandwidth1 80kbps
Deadline 60 seconds
• Receiver’s bandwidth
Size 250kbs
For each scenario, we should compare the results of each Available 2
algorithm according to the following criteria:
• Number of missed segments 5.2 Impact of missed Segments:
• Running time The Graph here shows the segment loss is less in load
In the current evaluation, the maximum receiver bandwidth sharing scheduling algorithm as compared to other dynamic
constraint has been ignored for all test cases. Since we are algorithms. Measuring the available bandwidth is of great
most interested in the effects of the bit rate on the number of importance for predicting the end-to-end performance of
missed segments and the algorithm execution time, most applications, for dynamic path selection and traffic
results provide an execution time or number of missed engineering, and for selecting between a numbers of
segments vs. bit rate graphs. More specifically, we plot the differentiated classes of service. The available bandwidth is
following result graphs: an important metric for several applications, such as grid,
a. Execution time/missed segments vs. time for varying video and voice streaming, overlay routing, p2p file
number of segments transfers, server selection, and inter domain path
b. Execution time/missed segments vs. time for varying monitoring. The overall throughput improves if all the
number of senders nodes seek to improve the diversity of segments in the
c. Execution time/missed segments vs. time for a varying network.
average sender bandwidth.
d. Missed segment and execution time algorithm
comparison graph.
On the other hand, increasing the number of segments
(video length) has little effect on the number of missed
segments or at least little effect on the bit rate at which the
number of missed segments starts to increases but in the
case of the segment scheduling algorithm, longer videos
take more time to schedule. This could possibly be caused by
the fact that those 2 algorithms are much faster than the
Load Sharing Scheduling algorithm and the inputs provided Graph1. Bit Rate v/s Time.
in the tests do not stress the algorithms enough to product From the above graph we can conclude that as the
trends in the execution time graphs. transmission time increases for the senders the amount of
When comparing the algorithms, the general trends data that can be sent through the sender of a particular
exhibited by all 3 of them are very similar. When comparing segment, the data rate of the segment decreases.
actual values, it becomes evident that the Load Sharing
Scheduling algorithm is superior to the other 2 when it
comes to minimizing the number of missed segments but in
most cases, this algorithm takes a lot to compute the results.
This table-1 summarizes various system parameters
which are included in the simulation calculation and the
system used for our experiments. Before presenting
empirical results, we present the simulation model as
follows. Table 1 summarizes the configuration parameters of
Graph 2. Time v/s Bit rate for varying Segments
(IJCNS) International Journal of Computer and Network Security, 85
Vol. 2, No. 9, September 2010

From the above graph we can conclude that as transmission locally, then the user will suffer a (moderate) waiting time
time increases for the senders the number of segments that it as the system searches for segments of peers to download the
can handle in any given time reduces and hence the bit rate desired content from.
decreases.
5.3 Scalability:
Generally speaking, scalability can be defined as the References
adaptability to changes in the peer-to-peer system size, load [1] Gnutella, “http://gnutella.wego.com/”.
extent and nature of the load. That is, the network load [2] Y.-H. Chu, S. G. Rao, and H. Zhang, “A case for end
should be distributed evenly among the peers, which means system multicast,” in Measurement and Modeling of
that every peer should be aware of approximately the same Computer Systems, 2000.
number of other peers. From the data it can be concluded [3] B. Cohen, “Incentives Build Robustness in BitTorrent,”
that our algorithm is more scalable than other dynamic in Workshop on Economics of Peer-to-Peer Systems,
algorithm such as segment scheduling and network coding. 2003.
The second approach to deal with the scalability issue of [4] V. Agarwal and R. Rejaie. Adaptive Multi-source
video streaming systems is to use P2P load sharing. “P2P Streaming in Heterogeneous Peer-to-Peer Networks In
networking architectures receive a lot of attention nowadays, MMCN, 2005.
as they enable a variety of new applications that can take [5] Y. Huang, Tom Z.J. Fu, Dah-Ming Chiu, J.C.S. Lui, and
advantage of the distributed storage and increased C. Huang, “Challenges, design and analysis of a large-
computing resources offered by such networks”. Their scale p2p-vod system,” Proc ACM SIGCOMM 2008, pp.
advantage resides in their ability for self organization, 375-388, 2008.
bandwidth scalability, and network path redundancy, which [6] X. Zhang, G. Neglia, J. Kurose, and D. Towsley. On the
are all very attractive features for effective delivery of media benefits of random linear coding for unicast applications
streams over networks .Also the space is saved for the peers in disruption tolerant networks. Second Workshop on
as the replicas will be deleted. Network Coding,Theory, and Applications (NETCOD),
2006.
6. Conclusion
[7] Wikipedia, P2P http://en.wikipedia.org/wiki/P2P,
The success of the peer-to-peer paradigm in both file accessed 20/11/2006.
distribution and live streaming application derived in the [8] M. R. Garey and D. J. Johnson, Computers and
adoption of this technology for the delivery of video-on- intractability: a guide to the theory of NP-completeness.
demand content. Fast response time is a technology factor New York: W.H. Freeman, 1979.
that end-users demand. Considerable research has been [9] C. Gkantsidis, J. Miller, and P. Rodriguez,
performed to find better ways to arrange data such that fast “Comprehensive view of a live network coding P2P
response time can be achieved by increasing throughput and system”, in Proc. ACM SIGCOMM/USENIX IMC’06,
maximum bandwidth with low startup delay. Load sharing Brasil, October 2006.
scheduling is better for improving the performance of peer- [10] S. Deering and D. Cheriton, “Multicast routing in
to-peer systems. Though further work is required towards a datagram internetworks and extended LANs”, ACM
better understanding of the efficacy of our algorithms in Transaction on Computer Systems, vo. 8, no. 2, pp. 85-
more realistic scenarios, we believe that the guidelines 110, May 1990.
proposed in this paper can be used to build high- [11] S. Banerjee, B. Bhattacharjee, and C. Kommareddy,
performance P2P VoD systems. Thus, high quality VoD is “Scalable application layer multicast”, in Proc. ACM
feasible with high playback rates. SIGCOMM’02, Pittsburgh, PA, August 2002.
Our system was designed to guarantee that the video [12] M. Castro, P. Druschel, A.-M. Kermarrec, A. Nandi,
starts playing shortly after the beginning of the download, A. Rowstron and A. Singh, “SplitStream: High-
and progresses without interruptions until the end of the bandwidth multicast in cooperative environments”, in
movie. While we have made an implicit assumption that Proc. ACM SOSP’03, New York, USA, October 2003.
users watch the entire video linearly, we believe that the [13] S. Acendanski, S. Deb, M. Medard, and R. Koetter,
same principles used in our system could be extended to “How good is random linear coding based distributed
support non-linear viewing, i.e., where users would be able networked storage?,” in NetCod, 2005.
to start watching from arbitrary points in the video and [14] PPLive internet site. http://www.pplive.com.
perform fast forward and rewind operations. However, if the [15] Xiaojun Hei, Chao Liang, Jian Liang, Yong Liu and
user desires to watch a part of the video that is not available Keith Ross, "Insight into PPLive: A Measurement Study
86 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 9, September 2010

of a Large-Scale P2P IPTV System", Proc of WWW


2006 workshop of IPTV services over World Wide Web.
[16] “PeerSim P2P Simulator,” 2005, accessed 01-May-
2006. [Online]. Available:
http://peersim.sourceforge.net/.
[17] [Agr02] R. Agrawal and V. Subramanian, “Optimality
of Certain Channel Aware Scheduling Policies,” Proc. of
2002 Allerton Conference on Communication,Control
and Computing, Oct. 2002.
[18] J.A. Pouwelse, J.R. Taal, R.L. Langendijk, D.H.J.
Epema,, “Real-time Video Delivery using Peer-to-Peer
Barterin Networks and Multiple
DescriptionCoding”http://ishare.ewi.tudelft.nl/public/Po
uwelseTaal_SMC2 004.pdf.
[19] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung,
“Network information flow,”
[20]H. Kopetz. Real-time systems-design principles for
distributed embedded applications. Kluwer Academic
Publishers, 1997.

Authors Profile

S.Nagaprasad Sri. Received the Bacholar


of Computer Applications (B.C.A.) From
Osmania University, Hyderabad, in 2001,
M.Sc (I.T.) From Sikkim Manipal
University, Manipal in 2003 and M.Tech
(C.S.) From Allahabad Agricultural
University, Allahabad in 2005. Presently
Pursing Ph.D in Computer Science &
Engineering from Aacharya Nagarjuna University, Guntur.

Prof. A.Vinay Babu obtained his


Bachelors degree in Electronics &
Communication Engineering from Osmania
University. Has duel Masters degree, one in
Computer Science & Engg and the other
one is in ECE from JNTU. He obtained his
PhD from JNTU, Hyderabad. His research
area is Data Mining and Image processing.

Prof. K.Madhukar received his Masters


in Engineering Physics and Instrumentation
from Osmania University in 1980 and PhD
from Osmania University in 1989.
Presently he is Professor and Head Dept of
Physics and Electronics, Nizam College,
Osmania University, Hyderabad. He
worked as ICT4D programme specialist in
United Nations Development Programme
(UNDP) for more than four years. His research area is Data
mining, Knowledge Discovery and Networks. Eight students
working for their doctorial degree under his supervision.

Potrebbero piacerti anche