An iterative multi-application mapping algorithm for heterogeneous MPSoCs

Abstract

Abstract-Task mapping plays a crucial role in achieving high performance and energy savings in heterogeneous multiprocessor platforms. The problem of optimally mapping tasks onto a set of given heterogeneous processors for maximal throughput/minimal overall energy consumption has been known, in general, to be NP-complete. This problem is exacerbated when mapping multiple applications onto the target platform. To address this problem, this paper proposes an iterative multi-application mapping algorithm that operates at run time. Based on statically derived optimal (or near optimal) mappings for each separate application, this algorithm will quickly find a near optimal mapping under the objectives of high performance and low energy consumption for the simultaneously running applications on heterogeneous platforms. We have evaluated our algorithm using a heterogeneous MPSoC system with three real applications. Experimental results reveal the effectiveness of our proposed algorithm by comparing derived solutions to the ones obtained from other well-known algorithms.

A hierarchical run-time adaptive resource allocation framework for large-scale MPSoC systems

“…It means that multiple tasks (scheduled by FCFS policy) can be mapped onto the same processor. As each tile has the same architecture in our target MPSoC system, compared with the approach used in [25]. We can further simplify the design-time mapping optimising problem by considering only a partial target architecture (i.e., a tile) to limit the number of processing elements in the mapping problem.…”

“…In our SHARA framework, we address the complex task mapping problem on our tilebased heterogeneous MPSoC system using the idea of a hybrid task mapping technique proposed in our previous work [25] which prepares partial task mappings for workload scenarios at design time and completes the mappings for the entire scenario at run time. This task mapping approach was proposed for a small-scale heterogeneous MPSoC system.…”

“…After the GM finished the workload scheduling, the LM in the tile where the new workload has to be executed will further optimise the task mapping derived from the workload partition heuristic in the GM. In each LM, we adopt the EIM algorithm proposed in [25] without considering the energy constraint which generates mappings with good quality in system throughput to further optimise the mapping for the new workload. The pseudocode of this algorithm is outlined in Algorithm 2.…”

“…This framework takes advantages of the state-of-the-art solutions for modern adaptive heterogeneous MPSoC systems where the system adaptivity is achieved by adaptively adjusting the mapping of applications to the underlying hardware resources which is optimised at design time. More specifically, we reconsider the method in the state-of-the-art solutions that allows for dynamically reconfiguring the system at run time based on pre-optimized system configurations, such as task mappings derived at design time [17,25,27,29,31,37,45], and extend it by solving the issues of scalability as introduced in [25] and blind adaptivity (a system reconfiguration that should not have happened because of its large overhead actually happened) that are usually existed in these solutions for our target large-scale heterogeneous MPSoC system.…”

Hardware/Software Codesign of Embedded Systems with Reconfigurable and Heterogeneous Platforms

“…+ 10], [hKYS + 12], [QP13a]). The authors of [JLK + 14] present a hybrid approach that performs compile-time scheduling and then chooses between the stored schedules at run-time.…”