05-09-2014, 10:53 AM
An Adaptive Cloud Downloading Service
[attachment=67656]
I. INTRODUCTION
Video content distribution is a challenging research problem because of its high bandwidth requirement
and the fast growing video population. In recent years, it is reported that Internet traffic is already
dominated by video [1]
II. ASSUMPTION
To keep the problem tractable, we make some simplifying assumptions for the models we study in
this paper.
• The set of videos remains unchanged. The study of video population and popularity churn is for
future work.
• Peer population is much larger than video population [10].
• All videos have the same size. The cloud storage can only store a small subset of the videos. In
simulation, we do use heterogeneous videos length obtained from practical system [6].
• Peers have the same upload capacity. In most networks, peer’s download capacity is much more
than upload capacity. We do not consider the constraint of download capacity in this work.
• Each peer issues one downloading request at any time. A multiple concurrent request model will be
considered in future study.
• Peers are fully connected, forming a full mesh topology [10]–[12].
• The cloud server is able to replace any cached video with a new one instantly. In other words, we
assume the time for video replacement can be ignored.
• The video size is large enough to be divided into an infinite number of small chunks. Based on this
assumption, we build a fluid model [11], [12].
We assume homogeneous upload capacity to simplify the scheduling strategy. In this paper, each peer
share its upload capacity to all other peers downloading the same video equally. In heterogeneous network,
peers with different upload capacity can be treated differently. Downloading performance can be affected
significantly by using different scheduling strategies. Intuitively speaking, other peers be benefit from
ADAPTIVE ALGORITHM
From the above analysis, there are both strengths and drawbacks for both the helper mode and server
mode. The helper mode wastes P2P resource because the cloud server needs to keep downloading new
content to help peers; while the server mode wastes the bandwidth resource of blocked peers. In this
section, we design an adaptive algorithm to determine the service mode for each movie. The cloud server
adjusts its strategy periodically, by running the following Automatic Mode Selection (AMS) algorithm to
determine the mode for each movie. We assume the value of N′
is known. The movies in helper mode
have higher priority to be included into cloud storage. Then, we consider the other movies in the order
of decreasing peer population.
CONCLUSION
In this work, we build a theoretical model to analyze different strategies for a cloud downloading system.
In particular, helper mode and server mode are used as abstraction of two different design philosophies
- using the cloud as peer or as server. Our analysis reveals that each strategy can be advantages, for
certain operating scenarios. Helper mode wastes some server bandwidth, but is best at leveraging P2P
capacity when request load is high. On the other hand, server mode is most efficient for dealing with large
video population relative to the cache size. We design an automatic mode selection (AMS) algorithm to
choose the suitable service mode for different scenarios. Our analysis helps a cloud downloading system
to optimize its design. We also discuss the potential benefit to apply our result for the mobile P2P case
[attachment=67656]
I. INTRODUCTION
Video content distribution is a challenging research problem because of its high bandwidth requirement
and the fast growing video population. In recent years, it is reported that Internet traffic is already
dominated by video [1]
II. ASSUMPTION
To keep the problem tractable, we make some simplifying assumptions for the models we study in
this paper.
• The set of videos remains unchanged. The study of video population and popularity churn is for
future work.
• Peer population is much larger than video population [10].
• All videos have the same size. The cloud storage can only store a small subset of the videos. In
simulation, we do use heterogeneous videos length obtained from practical system [6].
• Peers have the same upload capacity. In most networks, peer’s download capacity is much more
than upload capacity. We do not consider the constraint of download capacity in this work.
• Each peer issues one downloading request at any time. A multiple concurrent request model will be
considered in future study.
• Peers are fully connected, forming a full mesh topology [10]–[12].
• The cloud server is able to replace any cached video with a new one instantly. In other words, we
assume the time for video replacement can be ignored.
• The video size is large enough to be divided into an infinite number of small chunks. Based on this
assumption, we build a fluid model [11], [12].
We assume homogeneous upload capacity to simplify the scheduling strategy. In this paper, each peer
share its upload capacity to all other peers downloading the same video equally. In heterogeneous network,
peers with different upload capacity can be treated differently. Downloading performance can be affected
significantly by using different scheduling strategies. Intuitively speaking, other peers be benefit from
ADAPTIVE ALGORITHM
From the above analysis, there are both strengths and drawbacks for both the helper mode and server
mode. The helper mode wastes P2P resource because the cloud server needs to keep downloading new
content to help peers; while the server mode wastes the bandwidth resource of blocked peers. In this
section, we design an adaptive algorithm to determine the service mode for each movie. The cloud server
adjusts its strategy periodically, by running the following Automatic Mode Selection (AMS) algorithm to
determine the mode for each movie. We assume the value of N′
is known. The movies in helper mode
have higher priority to be included into cloud storage. Then, we consider the other movies in the order
of decreasing peer population.
CONCLUSION
In this work, we build a theoretical model to analyze different strategies for a cloud downloading system.
In particular, helper mode and server mode are used as abstraction of two different design philosophies
- using the cloud as peer or as server. Our analysis reveals that each strategy can be advantages, for
certain operating scenarios. Helper mode wastes some server bandwidth, but is best at leveraging P2P
capacity when request load is high. On the other hand, server mode is most efficient for dealing with large
video population relative to the cache size. We design an automatic mode selection (AMS) algorithm to
choose the suitable service mode for different scenarios. Our analysis helps a cloud downloading system
to optimize its design. We also discuss the potential benefit to apply our result for the mobile P2P case