Addressing the Influence of Hidden State on Wireless Network Optimizations using Performance Maps

Performance of wireless connectivity for network client devices is location dependent. It has been shown that it can be beneficial
to collect network performance metrics along with location information
to generate maps of the location dependent network performance. These
performance maps can be used to optimize the use of the wireless net-
work by predicting future network performance and scheduling the net-
work communication for certain applications on mobile devices. However,
other important factors influence the performance of the wireless communication such as changes in the propagation environment and resource
sharing.
In this work we extend the framework of performance maps for wireless
networks by introducing network state as an abstraction for all other factors than location that influence the performance. Since network state
might not always be directly observable the framework is extended with
a network state estimation function. To be able to take network state into
account when predicting network performance a network state prediction
function is also added. In our evaluation resource sharing is used as an
example of network state and the framework is applied to the use-case
of scheduling TCP-based communication to lower communication over-
head. We implement an ideal, a worst case and a probabilistically correct
realisation of the state estimation function and estimates its impact on
the map-based scheduling algorithm to show the full range of effects on
scheduling performance. Using extensive simulations for evaluation we
show how dynamic network state caused by resource sharing influences
the scheduling performance. Additionally we show that it is possible to
improve the application performance by including network state information in the communication scheduling compared to scheduling without
network state knowledge.