ATLAS: Adaptive Topology- and Load-Aware Scheduling

TL;DR

ATLAS is an adaptive scheduling protocol for wireless networks that quickly responds to topology and load changes, maintaining stable performance for multi-hop TCP flows.

Contribution

It introduces a distributed auction-based method for dynamic, topology- and load-aware scheduling without waiting for frame boundaries.

Findings

Adapts to network changes in less than 0.1 seconds

Maintains about 20% relative error in resource allocation

Supports TCP multi-hop flows in dynamic networks

Abstract

The largest strength of contention-based MAC protocols is simultaneously the largest weakness of their scheduled counterparts: the ability to adapt to changes in network conditions. For scheduling to be competitive in mobile wireless networks, continuous adaptation must be addressed. We propose ATLAS, an Adaptive Topology- and Load-Aware Scheduling protocol to address this problem. In ATLAS, each node employs a random schedule achieving its persistence, the fraction of time a node is permitted to transmit, that is computed in a topology and load dependent manner. A distributed auction (REACT) piggybacks offers and claims onto existing network traffic to compute a lexicographic max-min channel allocation. A node's persistence p is related to its allocation. Its schedule achieving p is updated where and when needed, without waiting for a frame boundary.We study how ATLAS adapts to controlled changes in topology and load. Our results show that ATLAS adapts to most network changes in less than 0.1s, with about 20% relative error, scaling with network size. We further study ATLAS in more dynamic networks showing that it keeps up with changes in topology and load sufficient for TCP to sustain multi-hop flows, a struggle in IEEE 802.11 networks. The stable performance of ATLAS supports the design of higher-layer services that inform, and are informed by, the underlying communication network.