Domain Project

We are excited to announce that our paper “Declarative Adaptive Optimization of Task-Based Applications on Heterogeneous Architectures” has been accepted at the WAMCA Workshop, part of the 37th International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD).

This work introduces a knowledge-based optimization framework that leverages Answer Set Programming (ASP) to dynamically allocate tasks in heterogeneous systems. By modeling applications through the Actor Model, the framework captures device constraints, workloads, and communication patterns, turning them into logical rules that guide runtime allocation decisions.

The proposed Actor Placement Optimizer (APO) adapts to changing workloads and heterogeneous hardware conditions, achieving near-optimal performance compared to an oracle with full knowledge. Key innovations include the introduction of multi-objective optimization metrics—computing-unit overload, inter-actor communication costs, and a novel annoyance measure that captures performance drags like rollbacks and synchronization delays.

Through extensive simulation experiments on CPUs, GPUs, and FPGAs, APO demonstrated performance improvements of up to 45% over traditional static partitioning and graph-based approaches, showing its effectiveness in maximizing throughput while avoiding costly migrations.

This result highlights the potential of declarative, model-driven optimization to support adaptive resource management in complex heterogeneous environments, reinforcing the project’s commitment to advancing high-performance computing methodologies. Moreover, this work bridges several of the components of the Domain project that have been published before, significantly moving our research results to a fully working prototype.