Decentralized dynamic resource management support for massively parallel processor arrays

Abstract

Abstract-This paper presents a hardware-supported resource management methodology for massively parallel processor arrays. It enables processing elements to autonomously explore resource availability in their neighborhood. To support resource exploration, we introduce specialized controllers, which can be attached to each of the processing elements. We propose different types of architectures for the exploration controller: fast FSMbased designs as well as flexible programmable controllers. These controllers allow to implement different distributed resource exploration strategies in order to enable parallel programs the exploration and reservation of available resources according to different application requirements. Hardware cost evaluations show that the cost of the simplest implementation of our programmable controller is comparable to our FSM-based implementations, while offering the flexibility for implementing different exploration strategies. We show that the proposed distributed approach can achieve a significant speedup in comparison with centralized resource exploration methods.

System integration of tightly-coupled processor arrays using reconfigurable buffer structures

“…The proposed architecture is especially suitable for the concept of Invasive Computing where an application can adapt itself to the number of available PEs in case of multiple applications competing about the PEs of the TCPA [9]. As an example, we have implemented a video processing demonstrator to perform edge detection on a video stream using a Laplace filter, which can process 1024×768 images at a rate of 60 frames per second.…”

Invasive Tightly-Coupled Processor Arrays

“…Based on many application requirements for 1D and 2D image, signal processing, and other compute-intensive loop specifications, the authors in Lari et al [2011] have proposed two major invasion strategies: linear invasion and rectangular invasion. Linear invasion, which results in capturing a linearly connected chain of PEs, is suitable for types of 1D digital signal processing applications.…”

“…The invaded neighbor, in turn, also continues the invasion in the same fashion. In Lari et al [2011], three different policies are proposed for claiming linearly connected regions of resources: (a) capturing of PEs in a sequence of straight lines of maximal length, (b) a strategy that chooses available neighbors in a random-walk fashion, and (c) a strategy that tries to have a meander-like capturing of PEs (for illustration see Figure 3(a)).…”

“…In Section 4, to some extent, our previous results [Lari et al 2011 are summarized and embedded in a larger context, respectively. However, most of the space in this article is devoted to the following new contributions.…”

“…To overcome this rigidity, our processor arrays support at hardware level the ideas of invasive computing such as resource exploration and management. For this purpose, we integrated new controllers in each Processor Element (PE) of a TCPA to enable extremely fast and decentralized resource allocation Lari et al [2011]. Additionally, these invasion controllers enable hierarchical power management in TCPAs .…”