Parallelising a model of bacterial interaction and evolution

Large simulations of bacterial colonies require huge amounts of computational time, the only way to achieve the necessary level of performance is with parallel computers and a suitably designed implementation that maps the problem onto the hardware. For real problems this mapping can be a non-trivial problem requiring careful consideration of the constraints in both the system being modelled and the hardware that executes that model. Here we describe an implementation of a system for modelling bacterial evolution that encompasses many physical scales. This system is composed entirely of individual entities all playing out a complex series of interactions. These individuals exist at the scale of the population of bacterial and at the gene product scale. This paper reports that it is possible to map a dynamic problem such as this onto fixed resources, for the most part making use of implicit multiplexing of resources provided by the OS and partitioning the problem to reduce communication time. Through this an efficient simulation can be created, making maximal use of the available hardware without constraining the model to require excessively specific resources.