John W. Hall

A Router Architecture for Real-Time Point-to-Point Networks

Parallel machines have the potential to satisfy the large computational demands of emerging real-time applications. These applications require a predictable communication network, where time-constrained traffic requires bounds on latency or throughput while good average performance suffices for best-effort packets. This paper presents a router architecture that tailors low-level routing, switching, arbitration and flow-control policies to the conflicting demands of each traffic class. The router implements deadline-based scheduling, with packet switching and table-driven multicast routing, to bound end-to-end delay for time-constrained traffic, while allowing best-effort traffic to capitalize on the low-latency routing and switching schemes common in modern parallel machines. To limit the cost of servicing time-constrained traffic, the router shares packet buffers and link-scheduling logic between the multiple output ports. Verilog simulations demonstrate that the design meets the performance goals of both traffic classes in a single-chip solution.

A router architecture for real-time communication in multicomputer networks

Parallel machines have the potential to satisfy the large computational demands of real-time applications. These applications require a predictable communication network, where time-constrained traffic requires bounds on throughput and latency, while good average performance suffices for best-effort packets. This paper presents a new router architecture that tailors low-level routing, switching, arbitration, flow-control, and deadlock-avoidance policies to the conflicting demands of each traffic class. The router implements bandwidth regulation and deadline-based scheduling, with packet switching and table-driven multicast routing, to bound end-to-end delay and buffer requirements for time-constrained traffic while allowing best-effort traffic to capitalize on the low-latency routing and switching schemes common in modern parallel machines. To limit the cost of servicing time-constrained traffic, the router includes a novel packet scheduler that shares link-scheduling logic across the multiple output ports, while masking the effects of dock rollover on the representation of packet eligibility times and deadlines. Using the Verilog hardware description language and the Epoch silicon compiler, we demonstrate that the router design meets the performance goals of both traffic classes in a single-chip solution. Verilog simulation experiments on a detailed timing model of the chip show how the implementation and performance properties of the packet scheduler scale over a range of architectural parameters.