Adaptive Backpressure: Efficient buffer management for on-chip networks

Abstract

Abstract-This paper introduces Adaptive Backpressure, a novel scheme that improves the utilization of dynamically managed router input buffers by continuously adjusting the stiffness of the flow control feedback loop in response to observed traffic conditions. Through a simple extension to the router's flow control mechanism, the proposed scheme heuristically limits the number of credits available to individual virtual channels based on estimated downstream congestion, aiming to minimize the amount of buffer space that is occupied unproductively. This leads to more efficient distribution of buffer space and improves isolation between multiple concurrently executing workloads with differing performance characteristics.Experimental results for a 64-node mesh network show that Adaptive Backpressure improves network stability, leading to an average 2.6× increase in throughput under heavy load across traffic patterns. In the presence of background traffic, the proposed scheme reduces zero-load latency by an average of 31 %. Finally, it mitigates the performance degradation encountered when latency-and throughput-optimized execution cores contend for network resources in a heterogeneous chip multi-processor; across a set of PARSEC benchmarks, we observe an average reduction in execution time of 34 %.

A detailed and flexible cycle-accurate Network-on-Chip simulator

“…BookSim2 furthermore supports several schemes for regulating buffer sharing in an effort to avoid performance pathologies that can result from allowing buffer space to be shared freely [6]. With credit-based flow control, such schemes can be implemented by restricting credit availability at the routers' output ports.…”

“…Several exemplary dynamic sharing policies are included with BookSim2; for example, one simple policy monitors the number of active VCs at each output and divides the credit pool evenly among them. A more advanced sharing policy tracks the level of the congestion for each active output VCs and allocates a larger share of buffer space to less congested VCs with the goal of avoiding inefficient buffer occupancy by blocked flits [6].…”

“…As a result, it has been widely used for research in many network contexts, including networks found in large-scale supercomputers and many-core processors. BookSim has been used to study many different aspects of network design, including topology [2], [3], routing [4], flow control [5], [6], router microarchitecture [7], quality-of-service [8], as well as new technologies such as nanophotonics [9], [10]. BookSim can also be incorporated into other system simulators to model the network; for example, GPGPU-sim [11], a many-core accelerator simulator for evaluating GPGPU workloads, leverages BookSim to model the on-chip communication.…”

Exploiting the Routing Efficiency for DMesh Networks

“…As known, routing algorithm [10][11][12][13], as well as buffer management [12], contributes to the performance of the whole network. In this paper, based on DMesh network, we analyze the unbalancing use of physical links in DXY algorithm, and present a novel routing algorithm to exploit routing adaptivity in order to improve performance of the overall chip.…”

A Buffered Flow Control Scheme with Flit Weight-Based Dropping Mechanism for Efficient Communication in NoC

“…Becker et al introduced in [13] a new scheme to control the utilization of router input buffers by the observation of the traffic. This mechanism heuristically limits credit availability for each virtual channel (VC) based on its observed performance characteristics.…”