RedScarf Reference Dataset

Published: 11-10-2019| Version 1 | DOI: 10.17632/wy9y7npwyv.1
Contributors:
Eduardo Alves da Silva,
Márcio Kreutz,
Cesar Zeferino

Description

This dataset comprises the results of an extensive set of simulation-based experiments that compare the performance of on-chip communication architectures. These experiments were performed using the RedScarf simulation environment, and the dataset includes raw data (simulation logs) and analyzed data (latency, jitter, and throughput) collected from the simulation of several communication architectures working under different traffic patterns. The dataset also includes the configuration files used to configure each experiment. The design space explored in these experiments includes: Traffic Scenarios used: • Uniform; • Matrix transpose; • Complement; • Butterfly; • Perfect Shuffle; and • Bit-reversal. Topologies evaluated: • 2D Mesh; • 3D Mesh; • 2D Torus; • Full Crossbar; • Bus emulated by an exclusive N-1-N (with only one exclusive channel) switched router; and • Chordal ring. Routing algorithms evaluated on a 2D Mesh: • XY; • Negative First; • North Last; • West First; and • Odd-even. Arbiter policies evaluated on a 2D Torus: • Static; • Rotative; • Random; and • Round-robin. Buffering dimensioning schemes evaluated on a 3D Mesh: • 4-flit input buffer; • 8-flit input buffer; • 16-flit input buffer; • 32-flit input buffer; • 4-flit input and 4-flit output buffers; • 8-flit input and 8-flit output buffers; and • 16-flit input and 16-flit output buffers. Metrics evaluated: • Latency (average, min, and max); • Accepted Traffic (throughput); and • Jitter (latency distribution).

Files

Steps to reproduce

Download RedScarf from GitHub (https://leds-lab.github.io/redscarf/) and install it (please follow the instructions available. After installing RedScarf, unzip the dataset file. It includes a set of directories for each experiment. To reproduce an experiment, open its configuration file (Configuration.rsf) in RedScarf, define the interval for analysis in the Performance Analysis tab, and select one of the alternatives to visualize the results.