1: \begin{abstract}

2:   We study probabilistic protocols for concurrent threshold-based load

3:   balancing in networks. There are $n$ resources or machines

4:   represented by nodes in an undirected graph and $m \gg n$ users

5:   that try to find an acceptable resource by moving along the edges of

6:   the graph. Users accept a resource if the load is below a

7:   \emph{threshold}. Such thresholds have an intuitive meaning, e.g.,

8:   as deadlines in a machine scheduling scenario, and they allow the

9:   design of protocols under strong locality constraints. When

10:   migration is partly controlled by resources and partly by users, our

11:   protocols obtain rapid convergence to a balanced state, in which all

12:   users are satisfied. We show that convergence is achieved in a number of 

13:   rounds that is only logarithmic in $m$ and polynomial in structural 

14:   properties of the graph. Even when migration is fully controlled by users, 

15:   we obtain similar results for convergence to approximately balanced states. 

16:   If we slightly adjust the migration probabilities in our protocol, we can 

17:   also obtain fast convergence to balanced states.

18: \end{abstract}

19: