Nice, France
December 16-18th, 2013
- Home
- Call for papers
- Committees
- Keynotes and tutorials
- Dates
- Submissions
- Accepted papers
- Contacts
- Venue & Accomodations
- Registration
- Program
Keynotes
- Marcos Aguilera, Microsoft Research, USA
- Geo-Distributed Storage in Data Centers: Data centers increasingly have a storage system that is {\em geo-distributed},
that is, distributed across several geographic locations.
We explain the general characteristics of this setting and the
challenges that it brings, chief among them the need to operate
with low latency despite significant network delays.
These challenges lead to many interesting problems:
migrating data online,
dealing with congestion,
providing efficient transactions, and more.
We discuss these problems and some recent solutions, which
bring together techniques from distributed computing,
distributed systems, and database systems.
Despite much progress, however, several algorithmic
and fundamental questions remain open and serve as
inspiration for further investigation.
- Geo-Distributed Storage in Data Centers: Data centers increasingly have a storage system that is {\em geo-distributed},
that is, distributed across several geographic locations.
We explain the general characteristics of this setting and the
challenges that it brings, chief among them the need to operate
with low latency despite significant network delays.
These challenges lead to many interesting problems:
migrating data online,
dealing with congestion,
providing efficient transactions, and more.
We discuss these problems and some recent solutions, which
bring together techniques from distributed computing,
distributed systems, and database systems.
Despite much progress, however, several algorithmic
and fundamental questions remain open and serve as
inspiration for further investigation.
- Eitan Altman, Inria, France
- Dynamic game models in complex systems: We begin the tutorial with a theoretic part that covers two areas: non-cooperative game theory, and population propagation models. In the game theory part, a particular attention will be given to potential games. We shall focus in particular on congestion games and on the game version of the generalized Kelly mechanism problem, both of which are known to be potential games. In our presentation of models for population propagation models, we shall present several models which we shall classify according to the size of population of potential interested destination nodes (which can be finite and constant, finite but non-constant or infinite), and the virality of the content. This will include branching and epidemic models. We shall then use these tools to study various applications to large networks. This will include (1) security issues related to e-virus attacks, (2) the question of what type of content should service providers specialis in, which will be solved by transforming it into an equivalent congestion game, (3) issues related to viral marketing and competition issues in social networks. In these problems the generalized Kelly mechanism will be frequently used. The game theoretic analysis will allow us to get insight on how much to spend on advertising products and on what product should we advertise.
Tutorials
- Hein Meling, University of Stavanger, Norway
- Paxos Explained from Scratch: (joint work with Leander Jehl) Paxos is a flexible and fault tolerant protocol for solving the consensus problem, where participants in a distributed system need to agree on a common value. However, Paxos is reputed for being difficult to understand. This tutorial aims to address this difficulty by visualizing Paxos in a completely new way. Starting from a naive solution and strong assumptions, Paxos is derived in a step-wise fashion. In each step, minimal changes are made to the solution and assumptions, aimed at understanding why the solution fails. In this manner, a correct solution that corresponds to Paxos is eventually reached.
- Marc Shapiro, Inria&LIP6, France, and Nuno Preguica, Universidade Nova de Lisboa, Portugal
- From strong to eventual consistency: getting it right: Distributed systems face a fundamental trade-off of consistency vs. availability and performance. Strong consistency is easy to understand but is slow, expensive, and is unavailable when the system partitions. Eventual consistency (EC) can be cheaper, faster, and more scalable, but is hard to understand and get correct. This tutorial explores the multiple gradations between strong and eventual consistency. It focuses on understanding EC, from perspectives of the algorithm designer, of the system builder, and the application programmer. It will include formal definitions of correctness, study of lower bounds, and implementation recipes and tricks.