This repository includes Caladan and several applications that were evaluated in the Caladan paper submitted to OSDI '20.
The main Caladan source code repository is linked as a submodule. It contains instructions for getting started with Caladan, and more details on supported platforms and use cases.
This repository contains scripts to facilitate building the applications used in the evaluation and running experiments from the paper.
This code was tested on Ubuntu 18.04 kernel version 5.2.0 on a server with a Mellanox ConnectX-5 40Gb/s NIC and an Intel Optane SSD.
Client machines require either an Intel 82599 based-NIC or a ConnectX-{3,4,5} Mellanox NIC.
First, install any needed packages:
sudo apt install build-essential libnuma-dev clang autoconf autotools-dev m4 automake libevent-dev libpcre++-dev libtool ragel libev-dev moreutils parallel cmake python3 python3-pip libjemalloc-dev libaio-dev libdb5.3++-dev numactl hwloc libmnl-dev libnl-3-dev libnl-route-3-dev uuid-dev libssl-dev libcunit1-dev pkg-config
Install rust, and use the nightly toolchain. See http://rust-lang.org/ for details.
curl https://sh.rustup.rs -sSf | sh
rustup default nightly
On the server machine, clone this repository in your home directory and build everything:
cd caladan-artifact
./build_all.sh
On each client machine, clone this repository in your home directory and build using:
cd caladan-artifact
./build_client.sh
The experiment scripts assume that binaries are located at the same absolute paths on client and server machines, so please use the same directory names.
This repository contains a collection of python scripts to execute experiments from the Caladan paper and to graph the results.
The experiment scripts require python3 and several libraries installed via pip3.
pip3 install matplotlib
pip3 install pandas
Each experiment run produces a folder with data collected from the run. To graph
the data, run python3 graph.py <folder1> <folder2>...
.
The scripts assume that passwordless SSH and passwordless sudo are setup on all the machines in the testbed.
This file describes the testing environment and should be modified to reflect your setup. We include the default setup for running on our testbed as an example. MACHINES
should contain information about each of the machines in the testbed, and CLIENT_SET
should contain the set of machines used to generate load.
Each machine is assumed to have (A) a management interface with its own address (oob_ip
) that is used for SSH connections and launching experiments and (B) a NIC used for the experiment.
To run the experiments, follow the above instructions to build and setup your environment. Run python3 paper_experiments.py
to run all the experiments. The total runtime depends on the configured number of samples and sample durations (found at the top of paper_experiments.py
).
After the experiments complete, run python3 graph.py run.2*
to generate graphs from each experiment (saved as pdfs).