Link to arXiv version

columbia · Jun 30, 2021 · 27d7532 · 27d7532
1 parent b1aa6ca
commit 27d7532
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 5 deletions.
diff --git a/README.md b/README.md
@@ -1,14 +1,14 @@
 
 # PrivateKube
 
-PrivateKube is an extension to the popular Kubernetes datacenter orchestrator that adds privacy as a new type of resource to be managed alongside other traditional compute resources, such as CPU, GPU, and memory.  A description of the project can be found on our [webpage](https://systems.cs.columbia.edu/PrivateKube/) and in our OSDI'21 paper, titled [Privacy Budget Scheduling](https://www.usenix.org/conference/osdi21/presentation/luo) (PDF locally available [here](https://columbia.github.io/PrivateKube/papers/osdi2021privatekube.pdf)).
+PrivateKube is an extension to the popular Kubernetes datacenter orchestrator that adds privacy as a new type of resource to be managed alongside other traditional compute resources, such as CPU, GPU, and memory.  A description of the project can be found on our [webpage](https://systems.cs.columbia.edu/PrivateKube/) and in our OSDI'21 paper, titled [Privacy Budget Scheduling](https://www.usenix.org/conference/osdi21/presentation/luo) (PDF locally available [here](https://columbia.github.io/PrivateKube/papers/osdi2021privatekube.pdf) and extended version available on [arXiv](https://arxiv.org/abs/2106.15335).
 
 
 ## Repo structure
 
 This repository contains the artifact release for the OSDI paper:
 - [system](system/): The PrivateKube system, which implements the privacy resource and a new scheduling algorithm for it, called *Dominant Privacy Fairness (DPF)*.
-- [privatekube](privatekube/): A python client for interaction with the PrivateKube system and performing macrobenchmark evaluation.
+- [privatekube](privatekube/): A Python client for interaction with the PrivateKube system and performing macrobenchmark evaluation.
 - [simulator](simulator/): A simulator for microbenchmarking privacy scheduling algorithms in tightly controlled settings.
 - [examples](examples/): Usage examples for various components, please refer its [README](./examples/README.md) for details.
 - [evaluation](evaluation/): Scripts to reproduce the macrobenchmark and microbenchmark evaluation results from our paper.
@@ -27,7 +27,7 @@ This repository contains the artifact release for the OSDI paper:
     - [1.4. Example usage in a DP ML pipeline](#14-example-usage-in-a-dp-ml-pipeline)
   - [2. Getting started with the simulator](#2-getting-started-with-the-simulator)
     - [2.1 Setup](#21-setup)
-      - [Setup python environment](#setup-python-environment)
+      - [Setup a Python environment](#setup-a-python-environment)
       - [Installation from source](#installation-from-source)
     - [2.2 Examples](#22-examples)
       - [The minimal simulation example](#the-minimal-simulation-example)
@@ -63,6 +63,7 @@ mkdir ~/.kube
 
 sudo chown -f -R $USER ~/.kube
 ```
+(You can learn more about how to use Microk8s without sudo [here](https://github.com/ubuntu/microk8s/blob/feature/dev-docs/docs/access-without-sudo.md))
 
 You can now start and stop your cluster with:
 ```bash
@@ -247,8 +248,8 @@ This simulator is used for prototyping and microbenchmark evaluation of privacy
 
 ### 2.1 Setup
 
-#### Setup python environment
-Install conda, create and activate an isolated python environment "ae". 
+#### Setup a Python environment
+Install Conda, create and activate an isolated Python environment "ae". 
 ```bash
 wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
 bash ~/miniconda.sh -b -p $HOME/miniconda

diff --git a/evaluation/macrobenchmark/README.md b/evaluation/macrobenchmark/README.md
@@ -1,14 +1,21 @@
 
 # Macrobenchmark
 
+
+## Requirements
+
 The commands in this section have to be run from the `macrobenchmark` directory. You can jump there with:
 
 ```bash
 cd evaluation/macrobenchmark
 ```
 
+You should have the `privatekube` Python package installed, as described in the [main README](https://github.com/columbia/PrivateKube).
+
 Please note that the steps below can take several days to run, depending on your hardware. If you want to examine the experimental data without running the preprocessing or the training yourself, you can download some artifacts from this [public bucket](https://storage.googleapis.com/privatekube-public-artifacts). 
 
+Training will be faster with a Nvidia GPU, but you can also use your CPU by specifying `--device=cpu` in the script arguments.
+
 ## Download and preprocess the dataset
 
 To download a preprocessed and truncated (140Mb instead of 7Gb) version of the dataset, run the following: