24-09-2014, 12:46 PM
Distributed Cooperative Caching
in Social Wireless Networks
Distributed Cooperative Caching.pdf (Size: 1.71 MB / Downloads: 37)
INTRODUCTION
1.1 Motivation
RECENT emergence of data enabled mobile devices and
wireless-enabled data applications have fostered new
content dissemination models in today’s mobile ecosystem.
A list of such devices includes Apple’s iPhone, Google’s
Android, Amazon’s Kindle, and electronic book readers
from other vendors. The array of data applications includes
electronic book and magazine readers and mobile phone
Apps. The level of proliferation of mobile applications is
indicated by the example fact that as of October 2010,
Apple’s App Store offered over 100,000 apps that are
downloadable by the smart phone users
User Selfishness
The potential for earning peer-to-peer rebate may promote
selfish behavior in some users. A selfish user is one that
deviates from the network-wide optimal policy in order to
earn more rebates. Any deviation from the optimal policy is
expected to incur higher network-wide provisioning cost. In
this work, we analyze the impacts of such selfish behavior
on object provisioning cost and the earned rebate within the
context of an SWNET. It is shown that beyond a threshold
selfish node population, the amount of per-node rebate for
the selfish users is lower than that for the nonselfish users.
In other words, when the selfish node population is beyond
a critical point, selfish behavior ceases to produce more
benefit from a rebate standpoint.
Search Model
After an object request is originated by a mobile device, it
first searches its local cache. If the local search fails, it
searches the object within its SWNET partition using limited
broadcast message. If the search in partition also fails, the
object is downloaded from the CP’s server using the CSP’s
3G/4G cellular network. In this paper, we have modeled
objects such as electronic books, music, etc., which are time
nonvarying, and therefore cache consistency is not a critical
issue. We first assume that all objects have the same size
and each node is able to store up to “C” different objects in
its cache. Later, in Section 5.3, we relax this assumption to
support objects with varying size. We also assume that all
objects are popularity-tagged by the CP’s server [3]. The
popularity-tag of an object indicates its global popularity; it
also indicates the probability that an arbitrary request in the
network is generated for this specific object.