Description

This is a supplementary material for the EVStore paper. This repo contains experiment that focuses on cache benchmarking experiments.

---

Run Cacheus Cache Benchmark

---

Follow: Cacheus Readme

---

Run Cache Benchmark on Local Machine

---

1. Uncompress the cache2k.zip

2. Put it in the maven repo, by default it should be at ~/.m2/repository/org/

3. The content of that cache2k should be here: ~/.m2/repository/org/cache2k/*

4. The main file is at cache-benchmark/clockProPlus/src/main/java/org/cache2k/benchmarks/clockProPlus/MainExperiment.java

   # Edit the following variables inside MainExperiment.java:
   
   BASE_DIR = "/Users/daniar/Documents/Github/cache-benchmark/inf-workload-traces/criteo_kaggle_all_mmap/inference=0.003/";
   BENCH_DIR = "caching_bench/";
   SUMMARY_DIR = "summary/";
   SUMMARY_OUTPUT_DIR = "caching_hit_summary/";
   
   benchWorkloadSize = 137521;
   totalCacheSize = 11854;
   policies = new String[] {  "EvLFU", "EvCAR", "EvARC" , "ClockPro"};

---

Run Cache Benchmark on Chameleon (CPU Node)

---

You can reserve any Skylake/Haswell/Cascadelake node on Chameleon Cloud

Create Reservation

Make sure the lease lasts for maximum 7 days through the lease menu

Launching an Instance

Use CC-Ubuntu20.04 as the node's operating system, for more information please refer to the Chameleon Cloud documentation

Allocate floating IPs

Allocate an IP address for the node so that it can be accessed through the Internet via SSH

1. Preparation [Login as "cc"]

• Use cc user

  ssh cc@<IP_address>     # use your IP address

• Check if it uses SSD

  lsblk -o name,rota      # 0 means SSD

• Setup disk
  Check if there is already mounted disk

  df -H
      # sda      8:0    0 223.6G  0 disk
      # should be enough

  Check SSD or Disk

  lsblk -d -o name,rota
  cat /sys/block/sda/queue/rotational

• Setup user

  sudo adduser --disabled-password --gecos "" evstoreuser
  sudo usermod -aG sudo evstoreuser
  sudo su 
  cp -r /home/cc/.ssh /home/evstoreuser
  chmod 700  /home/evstoreuser/.ssh
  chmod 644  /home/evstoreuser/.ssh/authorized_keys
  chown evstoreuser  /home/evstoreuser/.ssh
  chown evstoreuser  /home/evstoreuser/.ssh/authorized_keys
  echo "evstoreuser ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers.d/90-cloud-init-users
  exit
  exit

2. Setup zsh [Login on "user"]

ssh evstoreuser@<IP_address>

sudo su
apt-get update
apt-get install zsh -y
chsh -s /bin/zsh root
# Break the Copy here ====

exit
sudo chsh -s /bin/zsh evstoreuser
which zsh
echo $SHELL
sudo apt-get install wget git vim zsh -y
# Break the Copy here ====

printf 'Y' | sh -c "$(wget -O- https://raw.githubusercontent.com/ohmyzsh/ohmyzsh/master/tools/install.sh)"
/bin/cp ~/.oh-my-zsh/templates/zshrc.zsh-template ~/.zshrc
sudo sed -i 's|home/evstoreuser:/bin/bash|home/evstoreuser:/bin/zsh|g' /etc/passwd
sudo sed -i 's|ZSH_THEME="robbyrussell"|ZSH_THEME="risto"|g' ~/.zshrc
zsh
exit
exit

---

Run Cache Benchmark (EV-Store Java Version) on Chameleon

---

1. Clone the repo

cd ~
git clone https://github.com/daniarherikurniawan/cache-benchmark.git

2. Install dependencies

# Install the default Java Runtime Environment (JRE) and JDK 
cd ~/cache-benchmark		
sudo apt update
sudo apt install default-jre
sudo apt install default-jdk

# install maven
sudo apt install maven

mvn help:evaluate -Dexpression=settings.localRepository

# Make sure this folder already exists:
# ~/.m2/repository
cp ~/cache-benchmark/repository/* ~/.m2/repository/

3. Run cache-benchmark

• Preparation (get the workload) Follow: Set up Criteo Kaggle workloads for Java Cache-benchmark

• Do benchmark

  cd ~/cache-benchmark
  # Make sure the dataset path in MainExperiment.java matches yours  
  # MainExperiment.BASE_DIR = "/home/evstoreuser/workload/workload_5mill_kaggle/" in main function;
  chmod +x benchmark.sh
  
  # Change the parameters in MainExperiment.java at main() as needed, then run:
  
  cd ~/cache-benchmark
  ./benchmark.sh

---

Set up Criteo Kaggle workloads for Java Cache-benchmark

---

Converting raw Criteo Kaggle dataset to Java cache-benchmark workload

1. Get the raw dataset

mkdir ~/workload
cd ~/workload
wget http://go.criteo.net/criteo-research-kaggle-display-advertising-challenge-dataset.tar.gz
tar -xzvf criteo-research-kaggle-display-advertising-challenge-dataset.tar.gz
# Use partial dataset!
head -n 5000000 train.txt > train_5mill.csv

2. Convert raw dataset to Java cache-benchmark workload

# create output directory for new workloads
mkdir ~/workload/workload_5mill_kaggle

# Change inFile and outFile like yours 
# inFile : input path for raw dataset
# outDir : output directory for new workloads generated
python ~/cache-benchmark/script/converter.py --inFile ~/workload/train_5mill.csv --outDir ~/workload/workload_5mill_kaggle

Contributors

• Daniar Kurniawan
• Ruipu (Rex) Wang
• Fandi Wiranata
• Kahfi Zulkifli
• Garry Kuwanto

Acknowledgement

This code is based on:

• Cache2k (https://github.com/cache2k/cache2k)
• Cacheus (https://github.com/sylab/cacheus/)