On the use of multiple heterogeneous devices to speedup the execution of a computational model of the Human Immune System

The Human Immune System (HIS) is responsible for protecting the body against diseases, but the mechanisms used in this task are not completely understood. Mathematical and computational tools can be used for this purpose, and due to the costs involved in simulating the HIS, GPUs (Graphics Processing Units) are frequently used as the computational platform. The frequency in which GPUs are used for tasks like this is so high that some processing units, such as the APUs (Accelerated Processing Units), have integrated then into the CPU chip. This work presents the implementation on an APU of a mathematical model that describes part of the HIS. A load balancing strategy was implemented to distribute data with the objective of equalizing the load at each computational device, since GPU and CPU are heterogeneous. Gains up to 6.0×,1.28× and 3.7× were obtained by the balanced version of the code, when compared to the same parallel versions that execute exclusively on CPU, GPU, and on both of them, but without using load balancing, respectively.