Is There a Case for Mobile Phone Content Pre-staging? (1)

1. INTRODUCTION

Content caching is a key component of today's Internet
architecture. From large scale content delivery networks
to transparent caches and web proxies, caching has helped
make content delivery ecient, and has aided both end-users
and network providers of wired networks [1]. The last few
years have however witnessed a massive increase in content
accessed from cellular networks, where the bottlenecks are
dierent from traditional wireline networks. The \last mile"
of the cellular network is constrained by spectrum that needs
to be shared by multiple users. In this scenario, both for-
ward caching [2] and content pre-fetching [3] do not help
since an object traverses the last mile at least once for each
user requesting it. Hence, caching needs to be reassessed for
cellular networks.

While caching typically works as a \pull" based mecha-
nism, it can also support a \push" based mechanism, where
content is pushed/pre-staged to the cache before it is actu-
ally needed. In general, such an approach is not advocated
as it is JIAYU G4S hard to predict what content would be needed in
the future. However, recent works have proposed leveraging
external information to predict future accesses (e.g., from
social networks), or past information to pre-stage content
on devices [4, 5]. Pre-staging content is particularly suitable
for cellular networks, where the broadcast channel [6] can
be used to replicate on mobile devices popular content \for
free", saving network resources and possibly energy. What
is ZOPO ZP590 not known however is the potential benet that can be
expected by using such a solution.

The main contribution of this paper is to quantify the
potential gains, as well as overheads, of such a pre-staging
mechanism for content delivery over cellular networks. Us-
ing a large scale nation-wide anonymized HTTP trace from
actual ZOPO ZP590 web users where each request is annotated
with the geographical area where the requests originate, we
try to answer the following questions: Is the popularity of
content downloaded by mobile terminals skewed enough to
permit caching gains? Do content providers allow caching
of popular content? How large should the local cache be
for savings? What would be the savings in terms of traf-
c volume? And would the benets for end-users/content
providers be substantial enough for them to adopt such a
technology?

We rst analyse the content popularity nding that it fol-
lows the classic Zipf law, with most of the content having a
lifetime longer than one hour. Both properties are desirable
for ecient pre-staging. We then quantify possible gains
of the pre-staging system when proactively and periodically
broadcasting to mobile devices a \bundle"with the most ex-
pensive (popular and voluminous) content observed in past
periods. Using our traces, we show that such a system, if
present, would i) reduce the wireless link load by up to 40%
while ii) improving the end-users' performance by avoiding
the download of up to 20% of objects in optimistic scenarios
with a bundle of 100 MB.

However, performance of such an optimal system is depen-
dent on the bundle (and local Leagoo Lead 3) size. Indeed, bundling
the most expensive content would entail broadcasting pop-
ular elephants, creating bundles of several GBs. We further
discover that some popular objects are forced to be \not-
cacheable" by content providers, despite the objects being
static and very popular. Considering conservative scenarios
where the cache size is limited to 100 MB and only cacheable
objects can be bundled, overall savings reduce to about 10%
for both wireless link load and end-users' downloaded vol-
ume. Given our results, there could be network deployments
that could benet from content pre-staging on JIAYU G4S Phones, but
such a decision would have to be based on the perceived
value gain vs. associated overheads, and such a decision is
bound to vary across networks.http://cicimobile.shockup.com/2014/08/29/whos-calling-demographics-of-mobile-phone-use-in-rwanda/