Priority-Aware Near-Optimal Scheduling for Heterogeneous Multi-Core Systems with Specialized Accelerators

Zhuo Chen,Diana Marculescu

DOI: https://doi.org/10.48550/arXiv.1712.03246

2017-12-09

Abstract:To deliver high performance in power limited systems, architects have turned to using heterogeneous systems, either CPU+GPU or mixed CPU-hardware systems. However, in systems with different processor types and task affinities, scheduling tasks becomes more challenging than in homogeneous multi-core systems or systems without task affinities. The problem is even more complex when specialized accelerators and task priorities are included. In this paper, we provide a formal proof for the optimal scheduling policy for heterogeneous systems with arbitrary number of resource types, including specialized accelerators, independent of the task arrival rate, task size distribution, and resource processing order. We transform the optimal scheduling policy to a nonlinear integer optimization problem and propose a fast, near-optimal algorithm. An additional heuristic is proposed for the case of priority-aware scheduling. Our experimental results demonstrate that the proposed algorithm is only 0.3% from the optimal and superior to conventional scheduling policies.

Performance

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to achieve efficient task scheduling in heterogeneous multi - core systems, especially in the case of including dedicated accelerators and task priorities, in order to maximize system throughput. Specifically, the paper focuses on how to design an optimal or near - optimal scheduling strategy in systems with different processor types and task affinities, so that the system can provide high performance under power consumption limitations.

The main contributions of the paper include:
1. **Mathematically determining the optimal scheduling strategy**: For affinity - based heterogeneous systems (including general - purpose processors and dedicated accelerators), the paper for the first time mathematically determines the optimal scheduling strategy.
2. **Generality and practicality of the strategy**: The proposed optimal scheduling strategy is very general and practical because it has no special assumptions about task arrival rates, task size distributions, and resource processing sequences.
3. **Proposing a heuristic algorithm (MAP) for quickly solving approximate optimal solutions**: Applicable to systems with any number of resource types.
4. **Proposing the optimization problem of priority - aware scheduling for the first time**: And proposing a heuristic algorithm (MIS) to solve this problem.
5. **Verifying the effectiveness of the algorithms through extensive simulations**: The results show that the MAP algorithm is only 0.3% lower than the optimal solution, and in terms of handling priority requirements, the MIS algorithm is 46% better than the MAP algorithm.

Through theoretical analysis and experimental verification, the paper shows the methods for achieving efficient task scheduling in heterogeneous multi - core systems and solves the limitations of existing methods under dynamic workload changes and non - Markov assumptions.