Conquering MPSoC Complexity with Principles of aSelf-Aware Information Processing Factory - Towards Networked and Data-centric Platforms (IPF 2.0)

点击翻译按钮获取中文摘要

This DFG/NSF continuation project proposal expands the idea of an Information Processing Factory (IPF) towards distributed autonomous many-core systems (MPSoC), as well as towards systems-of-MPSoCs with varying connectivity, in cyber-physical applications and the Internet of Things. IPF represents a paradigm shift in platform design, moving long term, robust and interdependent system operation in the focus of platform design rather than the existing hardware/software component-centricity. Such a refocus is not only needed to prepare for growing hardware problems towards the end of Moore’s Law, but is a necessity to manage complexity when networked IT systems not only by far outnumber humans but when their complexity by far exceeds human perception. While such complexity growth of IT applications and services is visible to large parts of the society, and is publicly discussed, the complexity growth equally affects the platforms (computers, communication, storage, etc.) which enable such growing applications. Like a factory, the IPF permanently adapts to changing requirements, resource availability and health state, and coordinates logistics of IPF clusters. To conquer complexity through as much as possible distributed autonomy with as little centralized control as necessary, the approach is based on the principles of self-awareness and self-organization to coordinate MPSoCs in one or several devices. Particular attention is needed for coexisting critical and mixed criticality applications where performance guarantees (real-time, safety, …) and isolation are mandatory throughout the product lifetime. In phase 1, the foundation for organizing MPSoCs along the IPF paradigm were laid and evaluated by realistic examples. The requested phase 2 goes beyond individual MPSoCs, first towards locally interconnected systems, which are still controlled by a common plan, then towards dynamically cooperating clusters with emergent behavior, comparable to the interplay of several factories in global supply chains. Vehicle platooning, with distributed collaborative planning, verification and sensing shall be used as common research scenario. The German partners will focus on dynamic control of data-centric in-vehicle MPSoC platforms, with Dynamic-NUMA as the new “leitmotif” or method for organizing, directing, storing as well as hardware and software (object) caching of high volume and high bit-rate data streams. Thus, the performance and energy efficiency of distributed systems with numerous I/O, sensor sources, as well as sensor fusion will be improved while, at the same time, compliance with real-time and safety guarantees will be ensured. For experimental validation, a physical prototype of a vehicle network and corresponding simulation environments are available. An excellent exchange with the US partners, at personal and technical levels, has been established during project phase 1 and will be continued.