Introduction - Data Storage for Social Networks: A Socially Aware Approach - page 1

Chapter 1
Introduction
Evidenced by the success of Facebook with more than 800 million users, Twitter
with more than 200 million users, and other sites alike, online social networks
(OSNs) have become ubiquitous, offering novel ways for people to access infor-
mation and communicate with each other. Nielsen published stats in June 2010
[ 28 ] showing that three of the world's top ten popular brands online are social-
media related and, for the first time ever, social network or blog sites are visited by
75% of global consumers who go online. Among mobile users, social networking
would surpass voice as the most popular form of mobile communication by 2015,
according to Airwide Solutions [ 2 ].
The increasing popularity of social networking is undeniable, and so scalability
is an important issue for any OSN that wants to serve a large number of users.
There are two main ways to address scalability: vertical scaling and horizontal
scaling. While vertical scaling scales “up” the system by adding more hardware
resources to the existing servers, horizontal scaling scales “out” the system instead,
by adding commodity servers and partitioning the workload across these servers.
Vertical scaling is simple to manage, but horizontal scaling is more cost-effective
and avoids the single-point-of-failure and bottleneck problems. The latter has been
a de facto standard when it comes to managing data at massive scale for many
OSNs today.
In a distributed storage system, where the data is partitioned across a number of
servers, the data can also be replicated to provide a high degree of availability in
case of failures. While data partitioning and replication is a well-known problem in
the literature of distributed database systems [ 1 , 14 , 20 , 33 , 35 - 37 ], OSNs represent
a novel class of data systems. In OSNs, a data read for a user often requires
fetching the data of its neighbors in the social graph (e.g., friends' status messages in
Facebook or connections' updates in LinkedIn). We refer to this property as social
locality . Recent works [ 9 , 32 ] have shown that distributed queries of small data
records reduce performance compared to local queries. Therefore, social locality
should be taken into account when designing a distributed storage for OSNs.