HOWTO-simulate.md

  ________                    .__   __________.___  ________
 /  _____/___________  ______ |  |__\______   \   |/  _____/
/   \  __\_  __ \__  \ \____ \|  |  \|    |  _/   /   \  ___
\    \_\  \  | \// __ \|  |_> >   Y  \    |   \   \    \_\  \
 \______  /__|  (____  /   __/|___|  /______  /___|\______  /
        \/           \/|__|        \/       \/            \/

GraphBIG: How to simulate

Introduction

It is a common practice for architecture research to perform timing/functional simulations with benchmark workloads. To help architecture research, GraphBIG provides support for architectural simulations of both CPU and GPU GraphBIG workloads.

Hints

Before simulating the workloads directly, there are a few issues should be carefully considered.

Benchmark arguments

The default execution argument can be found in each benchmark's Makefile. Detailed explainaion of the benchmark arguments can be found here. Please note that it is desirable to use a smaller dataset because of the limitation of regular architecture simulators.

Section of interest

Each GraphBIG has a data-loading phase, in which the graph data is loaded from files and represented in an in-mem data structure. The data-loading phase usually is fairly long because of the intense disk IO operations. Thus, the simulation should skip the loading phase and focus only on the section of interest (when the graph is actually being processed). To help users get a more accurate starting point of simulations, GraphBIG includes simulation flags to indicate the section of interest. The flags are two dummy function calls:

• SIM_BEGIN: starting point of section of interest.

• SIM_END: ending point of section of interest.

To enable the flags, include "SIM=1" at compiling time (further details can be found here). Users can trigger the timing simulation (or trace generation for trace-driven simulators) when the function call named as "SIM_BEGIN" is catched.