Self-aware adaptation via implementation hot-swap for heterogeneous computing

Modern computing systems contain more and more processing units that are increasingly difficult to exploit; statically optimizing software for all hardware architectures and execution scenarios pose serious challenges. Self-aware adaptive computing systems are capable of adapting their behavior thousands of times per second to accomplish given goals despite living and working in an unpredictable environment whose condition can vary continually. Changing the behavior of a computing system may benefit a wide variety of fields, raging from the embedded world (e.g., smart phones) to the supercomputers world (e.g., clusters) and is particularly useful for meeting performance, power consumption, and resource consumption challenges. With this paper we show the impact of using self-aware adaptive applications running on heterogeneous computing systems featuring diverse processing units. The operating system will answer requests for functionalities by choosing at runtime the best suiting implementations. During the applications lifetime, their performances are monitored and, if necessary, active implementations are changed using a hot-swap mechanism. This work presents our vision for self-aware adaptive applications, focusing its attention on a hot-swap mechanism proving its effectiveness using a cryptographic secure hash algorithm executed on the diverse processing units of a heterogeneous computing system.