Introducing Virtual Accelerators to Decrease the Communication Overhead of an Artificial Hormone System for Task Allocation

The Artificial Hormone System (AHS) is a completely decentralized operation principle for a middleware which can be used to allocate tasks in a system of heterogeneous processing elements (PEs) or cores. Tasks are scheduled according to their suitability for the heterogeneous PEs, the current PE load and task cooperation. The AHS also provides properties like self-configuration, self-optimization and self-healing by task allocation. The AHS is able to guarantee real time bounds for such self-X-properties. In order to cluster cooperating tasks on neighboring PEs, the AHS uses accelerator hormones. If a task is taken on a PE, it spreads these accelerator hormones in the vicinity to attract cooperating tasks thus reducing the communication distances. However, spreading the accelerator hormones produces additional communication overhead which can be significant in larger systems. In this paper the concept of virtual accelerators is introduced to completely avoid the overhead. Virtual accelerators are calculated locally instead of sending physical accelerators via the network. This can be achieved by exploiting the relationship between accelerators and other hormones. We show that virtual accelerators do neither change task clustering properties nor real-time behavior of the AHS, but are able to completely avoid the corresponding communication overhead.