Professional Documents
Culture Documents
Developing Regions
ABSTRACT results in long user idle times when browsing the web, mak-
This paper presents a time shifted web caching system named ing users less productive[1]. However during the night these
Bassa. It is capable of scheduling, fetching and caching of networks are not used at all, a pattern that has been ob-
web objects larger than a given size threshold to take place served in several countries[3]. We have observed the same
within the off peak time of the network. We have achieved pattern in university networks in Sri Lanka[7]. In these net-
promising results using this system by maximizing the use works both over utilization and under utilization are un-
of available bandwidth for more interactive web objects dur- desirable as bandwidth is an expensive commodity in the
ing peak hours while utilizing previously underutilized band- developing world. Over utilization of bandwidth makes the
width in off peak hours. By empirically analyzing the effect network congested during the daytime while under utiliza-
of time shifted caching on users we have unearthed that such tion wastes bandwidth during the night. As a solution, time
systems encourage them to download larger files more than shifted caching has been proposed by [3] and implemented
they did previously while accessing websites with more mul- by the authors [14]. These systems are capable of reduc-
timedia content such as falsh videos during the peak hours. ing the congestion during the daytime by shifting some of
In this paper we present those findings empirically based the large file downloads to the nighttime, which in turn will
on data gathered from proxy server logs at University of utilize abundant bandwidth.
Colombo School of Computing. Despite slow download speeds in developing regions users
download large video files, ISO images, binary installers that
take hours to get downloaded. In some cases large files will
Categories and Subject Descriptors be only available the next day morning so that users will
C.2.m [Computer-Communication Networks]: Miscel- have to keep their computers running through the night in
laneous their absence. Today the world wide web has become more
interactive by depending on a large number of small ob-
jects (in most cases objects smaller than a few mega bytes),
General Terms but bandwidth available for these small objects is adversely
Measurements, Performance affected by large file downloads that will take hours to com-
plete in any case. Once these small objects which are prone
Keywords to interaction are obstructed the overall responsiveness of
the web deteriorates. This paper empirically analyze the
Web Caching, Bandwidth, Developing Regions impact of different content types and sizes on daytime web
traffic to design and evaluate a time shifted web caching
1. INTRODUCTION system called Bassa.
Web caching has been the centerpiece in improving web From here onwards this paper is organized as follows. Sec-
performance for more than a decade since its inception. In tion 2 will go through related works with respect to web
previous studies web caches have shown varying hit rates in caching in developing regions. Section 3 will analyze web
developing regions such as 10%-25% in India [1] and nearly traffic traces gathered from proxy server logs to understand
50% in Cambodia[3]. Even with web caching, object down- the temporal behavior and object size distribution of web
load times can be quite high when accessing contents through traffic with respect to time shifted caching. Section 4 will
a low bandwidth link shared between multiple users [1]. This give details on design of Bassa and evaluate the effect of it.
is a quite a common situation in developing regions which Finally in section 5 we will conclude the work presented here
with details on remaining issues and future directions.
Permission to make digital or hard copies of all or part of this work for
2. RELATED WORK
personal or classroom use is granted without fee provided that copies are Analyses of web traffic done on traces collected from busy
not made or distributed for profit or commercial advantage and that copies proxy servers in developed world are plentiful. However
bear this notice and the full citation on the first page. To copy otherwise, to there have been quite a few such studies on traces gath-
republish, to post on servers or to redistribute to lists, requires prior specific ered from proxy servers hosted or in contact with developing
permission and/or a fee.
NSDR’11, June 28, 2011, Bethesda, Maryland, USA. regions[4, 3] which have been helpful in our work. Analy-
Copyright 2011 ACM 978-1-4503-0739-0/11/06 ...$10.00. sis done in [3] reveals the possibility of using time shifted
1
Requests (Peak)
Table 1: Details on Deployment Environment. Bandwidth (Peak)
Requests (Off Peak)
Number of Requests Data Volume in GB 0.1
Bandwidth (Off Peak)
Percentage
0.01
0.001
type of web search over email. Also there have been studies
Amount of Data
related to web interaction models in developing regions[1]. 0.6
1000
100000 100000
Frequency
Frequency
Frequency
Frequency
10000
10000 10000
100 1000
1000 1000
100
100 100
10
10 10 10
1 1 1 1
1 10 100 1000 10000 100000 1e+06 1 10 100 1000 10000 100000 1e+06 1e+07 1 10 100 1000 10000 100000 1e+06 1e+07 1 10 100 1000 10000 100000 1e+06 1e+07
Object Size Object Size Object Size Object Size
(a) Video Content (b) Application Content (c) Splitted by Time (d) All Objects
0.7
according to the previous calculations done using the fitted Data for Peak Hours
from 7.00 hours to 20.00 hours (peak region), while idle area
expands from 20.00 hours onwards (off peak region). These 0.4
Percentage
number of small objects in both video and application con-
tent types that will not get affected by the system. This is 0.01
highly desirable as users will get these objects, which are es-
sential to proper functionality of websites with rich content,
0.001
immediately.
0
0 50 100 150
Object Size Threshold
200 250 300
6. REFERENCES
[1] J. Chen, S. Amershi, A. Dhananjay, and
Figure 7: Bandwidth utilization vs lower bound of L. Subramanian. Comparing web interaction models
object size after time shifting. in developing regions. In Proceedings of the First ACM
1
Symposium on Computing for Development, ACM
0MB
70MB
DEV ’10, pages 6:1–6:9, New York, NY, USA, 2010.
0.8
ACM.
[2] J. Chen, L. Subramanian, and J. Li. Ruralcafe: web
Amount of Data