# Mittagsseminar (in cooperation with M. Ghaffari, A. Steger and B. Sudakov)

__Mittagsseminar Talk Information__ | |

**Date and Time**: Thursday, September 15, 2016, 12:15 pm

**Duration**: 30 minutes

**Location**: CAB G11

**Speaker**: Przemek Uznański

## Round-Robin Token Propagation: Additive Combinatorics Meets Distributed Load Balancing

We revisit the notion of load balancing schemes with local token propagation. We investigate the setting when the number of tokens is small when compared to the graph size, that is $k \ll m$. In such a scenario of low density of tokens, it is natural to expect that in addition to loads being well balanced across the network, the cumulated loads (loads summed over fixed, arbitrarily long, intervals of time) are also balanced.

In this setting we analyze a class of deterministic schemes. The starting point is round-robin token propagation (also known as the rotor-router model). This process, for a graph of $m$ edges with $k$ tokens, is known to reach in polynomial time a recurrent state, and in this work we exhibit new properties of this recurrent state.

For the considered class of processes, we obtain a series of additive bounds on the edge cumulated load discrepancy. Our results apply to specific graph classes and also show that, in general, the cumulated edge load discrepancy is $\widetilde O(1)$ whenever $gcd(k,m) = \widetilde O(1)$. In particular, this result holds always when the number of edges of the graph is a prime number.

Our results also yield as a corollary bounds on edge idle time of the process, i.e., the longest possible time between two consecutive appearances of a unit of load on an edge, taken over all edges. The task of circulating $k$ tokens to minimize edge idle time is a fundamental one in the study of the patrolling problem on networks. While the best centralized offline solution to the patrolling problem trivially achieves $\Theta(m/k)$ idle time, it is an open problem how to construct good distributed solutions. Our results on cumulative discrepancy automatically translate into a distributed solution to patrolling with $\widetilde O(1)$ competitive ratio on idle time, for all of the covered cases.

Upcoming talks | All previous talks | Talks by speaker | Upcoming talks in iCal format (beta version!)

Previous talks by year: 2018 2017 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996

Information for students and suggested topics for student talks

Automatic MiSe System Software Version 1.4803M | admin login