(Georgia Institute of Technology, 2011-01)
Shi, Cong; Lakafosis, Vasileios; Ammar, Mostafa H.; Zegura, Ellen W.
The opportunistic or disruption tolerant networking (DTN)
paradigm shows up in a variety of settings, from military to
disasters to the developing world to deep space; anywhere
that fixed infrastructure is either unavailable or expensive.
Simple messaging applications have substantial value for communication
and coordination. We posit that these settings
can also leverage applications that are computationally complex
and will benefit from distributed computing to take full
advantage of nearby computational resources. Computing
over these networks is not trivial, however, since network
disconnections are common and persist over many time scales.
In this paper we present the design and implementation of
Serendipity, a general purpose distributed computing platform
designed to run over a DTN. We have designed a simple
but powerful job structure that is suitable for such an underlying
network. As opposed to traditional distributed computing
platforms in data centers and clusters, where a central
master is used to allocate tasks and monitor the working
nodes, Serendipity relies on the collaboration among DTN
nodes on these functionalities. Smart task allocation algorithms
are designed to disseminate tasks among mobile devices
by accounting for the special properties of DTNs. The
extensive evaluation of our system on Emulab demonstrates
that Serendipity efficiently speeds up various kinds of distributed
computing jobs by a factor of 2.3 to 10.1 in a diverse
set of DTN environments.