Skip to content
This repository has been archived by the owner on Sep 12, 2024. It is now read-only.

Latest commit

 

History

History
205 lines (140 loc) · 6.97 KB

surfcursus-part-B-2015-Oct-15.markdown

File metadata and controls

205 lines (140 loc) · 6.97 KB
Raw
layout
default

SURFcursus - part B

This is part B of the Tutorial SURFcursus - 15 Oct 2015.
If you have not completed (and understood) SURFcursus - part A, please do so first.

3. Persistence

Disk images can be persistent or not and you can switch that flag.

  • What does persistent mean?

    • changes by the VM are copied back to the original disk image
    • if a VM runs with a persistent disk, you cannot launch a second VM using that disk

  • What does non-persistent mean?

    • changes by a VM are lost at shutdown and not copied back to the original disk image
    • it possible to run multiple VMs using the same non-persistent disk

Food for brain:

Was the first image that you imported persistent or not?

In this section you will work with persistent images to install an application in your VM. You will go through these steps:

  • Make your image persistent.
  • Deploy a VM using the persistent image.
  • Install and Run Mandelbrot in your VM.

Make your image persistent

  • Go to Images tab under Virtual Resources and click on the Cursus Image.
  • Under the Information section, find the Persistent entry. It says "no".
  • Switch the value to "yes".

Start a persistent VM

  • Start your virtual machine again.
  • Login and check the files you made during the previous run.

During the previous run the VM's disk was non-persistent.

Install and Run Mandelbrot

The disk image is now marked as persistent. This means that you can install your software or store your data in the VM and make sure that everything will remain in the VM once it is shutdown and re-instantiated.

Let's install and run an example to investigate the HPC Cloud capabilities. The code was taken from mandelbrot-rmi, a distributed Manderbrot renderer based on Java RMI, with some small improvements.

NOTE:

For the rest of the tutorial we will omit ubuntu@ip-145-100-59-102:~$ prompt in the instructions, in order to allow you copy-paste the commands directly in your terminal.

  • Login to the running VM again.

  • Install java as it is required for running the example (optionally try java -version to verify that is it is missing):

sudo apt-get install default-jre
  • Download the example:
wget https://doc.hpccloud.surfsara.nl/assets/mandelbrot-rmi-bin.tgz

Run the example

  • Create a blank directory and extract the tar:
mkdir mandelbrot-rmi
cd mandelbrot-rmi
tar xvzf ../mandelbrot-rmi-bin.tgz
ls
  • Launch the first Mandelbrot server that will pose as the RMI registry with an arbitrary name argument, e.g. apple. To do this:
    • Start a new local terminal (Linux or Mac users) or Gitbash (Windows) and log in to the same virtual machine.
    • Run the following commands:
cd $HOME/mandelbrot-rmi
./startserver.sh apple
  • Return to the first terminal window (connected to the VM with ssh -X). You will now start a Mandelbrot client and pass it the apple server name:
cd $HOME/mandelbrot-rmi
./startclient.sh apple

A window should pop up on your laptop displaying a full Mandelbrot set.

  • Interact with the display:

NOTE:

Left click will zoom in to the yellow rectangle.
Right click will zoom out.
Dragging will change the size of the rectangle.

Food for brain:

The Mandelbrot client and server both display logging info and the computation time. Zoom in on the mandelbrot and inspect this inforamtion.

You can stop the Mandelbrot client or server applications anytime with Ctrl+C.

  • Terminate the client now.

  • Launch additional Mandelbrot servers with different names. Start the new servers and the client, all in different terminals:

# new terminal, ssh to VM, then:
cd ~/mandelbrot-rmi
./startserver.sh banana

Repeat for coconut or make up your own names.

  • In the client window, start the client with the names of your running servers.
./startclient.sh apple banana coconut # or your names

Food for brain:

Observe the performance in the terminals when adding extra servers.

  • Before moving to the next exercise, remember to Shutdown your VM.

4. Scale-up to a multicore VM

The HPC Cloud is offered as an Infrastructure as a Service (IaaS). That allows you to give your Virtual Machines (VMs) the form that you need them to have. In this section, you will run the same Mandelbrot example on a four-core VM, using the exact same image, the Cursus Image. To scale-up your VM to use multiple cores you need to:

  • Edit your Template.
  • Re-instantiate a VM from the modified Template.

Edit the Template

You can customise your VMs by editing the templates you instantiate the VMs from.

  • In order to edit your existing template, choose the Cursus Template item and then Update buttons on the top right.
  • In the General tab edit the number of CPU and VCPU as:
    • CPU: 4
    • VCPU: 4
  • When you are done, click on the green Update button so that your changes are actually saved.

That was it. From now on you will get a 4-core VM running using the same Cursus Image (where the mandelbrot example is saved).

Run the example on the four-core VM

  • Instantiate a VM from the updated template.
  • Login to the VM. Is your mandelbrot data there?

NOTE: Your VM's image was (and is) persistent.

  • Run the same example with a single Mandelbrot server.
  • Start a new terminal and connect to the VM. Type the following command:
ps -eo pid,psr,pcpu,args | sed -n -e '1p' -e '/java/p'

This shows the running Java processes and the CPU number on which it runs (column PSR).

  • Stop Mandelbrot server(s) and client.
  • Launch multiple servers on different CPUs. To do this:
    • Start new terminals as needed and run variations on the following command:
cd $HOME/mandelbrot-rmi
taskset -c 2 ./startserver.sh mango # replace '2' with the CPU number and 'mango' with your server name
  • Start the client on a separate processor:
taskset -c 0 ./startclient.sh pineapple grape mango # use your server names. duh.

Food for brain:

Observe the performance. Is there any gain?

5. Wrap up

  • Play around, make your checks and don't forget to shut down all the running VMs when you are finished.

Remember: Your running VMs are consuming quota whether they are doing something useful or are idle.

Next: Extras

You have completed the Tutorial SURFcursus - part B. If you want more of the HPC Cloud, see Surfcursus - Extras.