Permalink
Branch: master
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
159 lines (97 sloc) 3.81 KB

ISCA 2018 Experiments

This page contains descriptions of the experiments in our ISCA 2018 paper and instructions for reproducing them on your own simulations.

One important difference between the configuration used in the ISCA 2018 paper and the open-source release of FireSim is that the ISCA paper used a proprietary L2 cache design that is not open-source. Instead, the open-source FireSim uses an LLC model that models the behavior of having an L2 cache as part of the memory model. Even with the LLC model, you should be able to see the same trends in these experiments, but exact numbers may vary.

Each section below describes the resources necessary to run the experiment. Some of these experiments require a large number of instances -- you should make sure you understand the resource requirements before you run one of the scripts.

Compatiblity: These were last tested with commit bba9dea4811a2445f22809ef226cf00971674758 of FireSim.

Prerequisites

These guides assume that you have previously followed the single-node/cluster-scale experiment guides in the FireSim documentation. Note that these are advanced experiments, not introductory tutorials.

Building Benchmark Binaries/Rootfses

We include scripts to automatically build all of the benchmark rootfs images that will be used below. To build them, make sure you have already run ./sw-manager.py -c br-disk.json build in firesim/sw/firesim-software, then run:

cd firesim/deploy/workloads/
make allpaper

Figure 5: Ping Latency vs. Configured Link Latency

Resource requirements:

To Run:

cd firesim/deploy/workloads/
./run-ping-latency.sh withlaunch

Figure 6: Network Bandwidth Saturation

Resource requirements:

To Run:

cd firesim/deploy/workloads/
./run-bw-test.sh withlaunch

Figure 7: Memcached QoS / Thread Imbalance

Resource requirements:

To Run:

cd firesim/deploy/workloads/
./run-memcached-thread-imbalance.sh withlaunch

Figure 8: Simulation Rate vs. Scale

Resource requirements:

To Run:

cd firesim/deploy/workloads/
./run-simperf-test-scale.sh withlaunch

Notes: Excludes supernode since it is still in beta and not merged on master.

Figure 9: Simulation Rate vs. Link Latency

Resource requirements:

To Run:

cd firesim/deploy/workloads/
./run-simperf-test-latency.sh withlaunch

Notes: Excludes supernode since it is still in beta and not merged on master.

Running all experiments at once

This script simply executes all of the above scripts in parallel. One caveat is that the bw-test script currently cannot run in parallel with the others, since it requires patching the switches. This will be resolved in a future release.

cd firesim/deploy/workloads/
./run-all.sh