PIO - Parallel IO

A framework for synchronized, concurrent, high-performance IO evaluation.

Setup

• Ensure you have Python 2.6 or 2.7 installed

    python --version
  

• Install Python setuptools

    # check if setuptools are installed
    python -c 'import setuptools'
  
    # if not then
    wget --no-check-certificate https://bootstrap.pypa.io/ez_setup.py    
    python ez_setup.py --insecure
  

• Install Cluster Shell

    wget http://sourceforge.net/projects/clustershell/files/clustershell/1.6/clustershell-1.6.tar.gz/download
    tar xvfz clustershell-1.6.tar.gz
    cd clustershell-1.6
    python setup.py install
    cp -r conf /etc/clustershell
  

• Setup no password SSH to all slave nodes (in this example they are named node01, node02 and node03):

    ssh-keygen -t rsa -N ''
    for n in node01 node02 node03; do
        ssh-copy-id -i ~/.ssh/id_rsa.pub $n
    done
  

• Install PIO (master node only):

    python setup.py install
  

• Install FIO on all slave nodes:

    # setup libAIO
    zypper install libaio libaio-devel          # sles
    yum install libaio libaio-devel             # redhat
    apt-get install libaio libaio-devel         # debian, ubuntu
  
    # install latest FIO version
    wget -O fio-master.zip https://github.com/axboe/fio/archive/master.zip
    unzip fio-master.zip
    cd fio-master
    ./configure
    make
    make install
  

• OPTIONAL Installing FSTEST - contact SAP for the following files: fstest, libhdbbasis.so, libhdblttbase.so, libhdbversion.so and run:

    mkdir -p lib/fstest
    cp fstest lib/fstest/
    cp libhdb*.so lib/fstest/
  

Usage

Performance Scenario

For example, a master node coordinates 3 slave nodes (node01, node02 and node03) to run several fio workloads that cover reads and writes, with different block sizes (4k, 16k, 64k, 256k, 1M), all random, where each job is run for 60 seconds.

pio perf -t random:read,write:4k,16k,64k,256k,1M -o output_file -l 60 \
    fio node01,node02,node03:/mnt/storage node02:/mnt/other-storage

• "/mnt/storage" is a directory that is accessible on slave nodes node01, node02 and node03 (does not have to be shared).
• "/mnt/other-storage" is a directory that is accessible on slave nodes node02.

In order to plot the data to a nice graph, run:

pio plot -a .
gnuplot plot.gp

Durability Scenario

Initialize a data file that lays on the evaluated storage system. Mess around with the system (power shortage, faulty disks, etc.), then check that the data is not corrupted.

pio dur init /mnt/storage

pio dur validate /mnt/storage