Reconfigurable Interconnection Network Laboratory

The mission

Our mission is to investigate reconfigurable interconnection networks and their applications in distributed computing and networks-on-chips. Reconfigurable interconnection networks are networks designed to be programmable in field for applications. We aim to provide reconfigurable interconnection network solutions, hardware/software designs, application oriented prototypes and products, and development tools.


The problem

Clusters, grids, clouds and Internet of Things provide rich and elastic computing resources for next generation of large and complex distributed computing applications. Such applications typically involve massive heterogeneous components running on a large number of networked devices known as nodes, and have high requirements on performance, security and fault-tolerance. As such, it is challenging to design, implement, test, deploy, and execute. We divide the challenge into two core problems: 1) reliable and efficient data communication among a large number distributed components, 2) scalability and fault-tolerance and automation in the design, deployment and execution management of heterogeneous components.


The solution

Our solution to the first problem is programmable net, a software-defined network that connects the nodes of a communication group in an application-specific topology optimized for communication efficiency. Net connections are established by programming nodes at run time. Net routing transfers data packets along programmable net to destination nodes. It pushes/pulls data to/from n nodes in time O(ln(n)) by optimal topology, allows relays to connect nodes in different domains, and supports reliable and secure data communications. The solution to the second problem is the reconfigurable component, a software module can be dynamically deployed, loaded, reconfigured and called. It allows to design a small number of prime components and then use them to build other components through combinations and reconfigurations. We created a unique software product called Distributed Application Platform (DAP) for the development, deployment and execution of distributed applications. DAP is a cross-platform software solution with both Java and C++ implementations and runs on major operating systems.


The root

Our root is in the interests of reconfigurable on-chip networks. The increasing number of heterogeneous components and complexity of inter-component communications are challenges in the design of system-on-a-chip. Packet-switched network architectures, i.e. Networks-on-Chip (NoC), have been considered to be the next generation on-chip network architectures in addition to bus and circuit-switched network architectures. We are particularly interested in packet-switched reconfigurable networks-on-chip architectures, which uses programmable net and net routing for inter-component communications. The core of DAP was originally developed as a software simulator for the packet-switched reconfigurable networks-on-chip architectures and turned out to be an effective solution for large scale distributed computing.

Copyright © 2024 RINLab@WLU