Expandable process networks to efficiently specify and explore task, data, and pipeline parallelism

Abstract

Running each application of a many-core system on an isolated (virtual) guest machine is a widely considered solution for performance and reliability issues. When a new application is started, the guest machine is assigned with an amount of computing resources that depends on the overall workload of the system and is not known to the designer at specification time. For instance, the computing resources might consist of many slow or a few fast processing elements. If the application is statically specified, as, for example, with Kahn process networks, the number of processing elements usable by an application is upper bounded by its number of processes. Similarly, the inter-process communication overhead might limit the maximum performance if the number of processing elements is significantly smaller than the number of processes. In this paper, we propose a formal extension for streaming programming models called expandable process networks (EPNs) that tackles this challenge by abstracting several possible granularities in a single specification. This enables the automatic exploration of task, data, and pipeline parallelism by two basic design transformation techniques, namely replication and unfolding. Then, the EPN semantics facilitates the synthesis of multiple design implementations that are all derived from one high-level specification. At runtime, the best fitting implementation for the given computing resources is selected to maximize the performance. Finally, we demonstrate the effectiveness of the proposed model on Intel's 48-core SCC processor.

Throughput-optimizing Compilation of Dataflow Applications for Multi-Cores using Quasi-Static Scheduling

“…In [17], expandable process networks are introduced. This work is orthogonal to our approach as it already assumes the presence of clusters which are functionally equivalent to a subgraph which is replaced by the cluster.…”

AdaPNet

“…In this paper, we represent an application as an EPN [17], i.e., the application is specified as a top-level process network, which can be refined by hierarchically replacing processes by other process networks. In this section, we formally specify the application model.…”

“…Specially, we address this challenge in the context of streaming applications, which are modeled as process networks. In fact, previous works showed that large performance gains in terms of throughput [17,20] or energy consumption [2] can be obtained if the process network is refined at compile-time.…”

“…Instead of adhering to a static programming model, thereby limiting the possible structural changes to the replication of stateless processes, we specify applications as expandable process networks (EPNs) [17]. EPNs extend conventional Kahn process networks (KPNs) [10] in the sense that several process networks with different degrees of parallelism are abstracted in a single specification.…”

Efficient Process Network Implementation of Ray-Tracing Application on Heterogeneous Multi-Core Systems

“…In our specific application, several copies of the intersect process can be instantiated and parallelized over multiple cores to take advantage of multi-core platforms. The Expandable Process Networks (EPN) MoC [10] is a model extension to KPN that aims to increase the degree of parallelism by properly expanding the topology of the original graph by means of replication and unfolding. Replication simply replicates a process many times while automatically creating all necessary channels and ports equivalent to the ones that were present in the original KPN specification.…”