7 min read

In this article by Carlos R. Morrison the authors of the book Build Supercomputers with Raspberry Pi 3, we will learn following topics:

  • Preparing master node
  • Transferring the code
  • Preparing slave node

(For more resources related to this topic, see here.)

Preparing master node

During boot-up, select the US or (whatever country you reside in) keyboard option located at the middle-bottom of the screen. After boot-up completes, start with the following figure. Click on Menu, then Preferences. Select Raspberry Pi Configuration, refer the following screenshot:

Menu options

The System tab appears (see following screenshot). Type in a suitable Hostname for your master Pi. Auto login is already checked:

System tab

Go ahead and change the password to your preference. Refer the following screenshot:

Change password

Select the Interfaces tab (see following screenshot). Click Enable on all options, especially Secure Shell (SSH), as you will be remotely logging into the Pi2 or Pi3 from your main PC. The other options are enabled for convenience, as you may be using the Pi2 or Pi3 in other projects outside of supercomputing:

Interface options

The next important step is to boost the processing speed from 900 MHz to 1000 MHz or 1 GHz (this option is only available on the Pi2). Click on the Performance tab (see following screenshot) and select High (1000MHz). You are indeed building a supercomputer, so you need to muster all the available computing muscle. Leave the GPU Memory as is (default). The author used a 32-GB SD card in the master Pi2, hence the 128 Mb default setting:

Performance, overclock option

Changing the processor clock speed requires reboot of the Pi2. Go ahead and click the Yes button. After reboot, the next step is to update and upgrade the Pi2 or Pi3 software. Go ahead and click on the Terminal icon located at the top left of the Pi2 or Pi3 monitor. Refer the following screenshot:

Terminal icon

After the terminal window appears, enter sudo apt-get update at the “$” prompt. Refer the following screenshot:

Terminal screen

This update process takes several minutes. After update concludes, enter sudo apt-get upgrade; again, this upgrade process takes several minutes. At the completion of the upgrade, enter at the $ prompt each of the following commands:

  • sudo apt-get install build-essential
  • sudo apt-get install manpages-dev
  • sudo apt-get install gfortran
  • sudo apt-get install nfs-common
  • sudo apt-get install nfs-kernel-server
  • sudo apt-get install vim
  • sudo apt-get install openmpi-bin
  • sudo apt-get install libopenmpi-dev
  • sudo apt-get install openmpi-doc
  • sudo apt-get install keychain
  • sudo apt-get install nmap

These updates and installs will allow you to edit (vim), and run Fortran, C, MPI codes, and allow you to manipulate and further configure your master Pi2 or Pi3. We will now transfer codes from the main PC to the master Pi2 or Pi3.

Transferring the code

IP address

The next step is to transfer your codes from the main PC to the master Pi2 or Pi3, after which you will again compile and run the codes, same as you did earlier. But before we proceed, you need to ascertain the IP address of the master Pi2 or Pi3. Go ahead and enter the command ifconfig, in the Pi terminal window:

The author’s Pi2 IP address is 192.168.0.9 (see second line of the displayed text). The MAC address is b8:27:eb:81:e5:7d (see first line of displayed text), and the net mask address is 255.255.255.0. Write down these numbers, as they will be needed later when configuring the Pis and switch.

Your IP address may be similar, possibly except for the last two highlighted numbers depicted previously. Return to your main PC, and list the contents in the code folder located in the Desktop directory; that is, type, and enter ls -la. The list of the folder content is displayed. You can now Secure File Transfer Protocol (SFTP) your codes to your master Pi. Note the example as shown in following screenshot. The author’s processor name is gamma in the Ubuntu/Linux environment:

If you are not in the correct code folder, change directory by typing cd Desktop, followed by the path to your code files. The author’s files are stored in the Book_b folder. The requisite files are highlighted in red. At the $ prompt, enter sftp [email protected], using, of course, your own Pi IP address following the @ character. You will be prompted for a password. Enter the password for your Pi. At the sftp> prompt, enter put MPI_08_b.c, again replacing the author’s file name with your own, if so desired. Go ahead and sftp the other files also. Next, enter Exit. You should now be back at your code folder.

Now, here comes the fun part. You will, from here onward, communicate remotely with your Pi from your main PC, same as the way hackers due to remote computers. So, go ahead now and log into your master Pi. Enter ssh [email protected] at the $ prompt, and enter your password. Now do a listing of the files in the home directory; enter ls -la, refer the following screenshot:

You should see the files (the files shown in the previous figure) in your home directory that you recently sftp over from your main PC. You can now roam around freely inside your Pi, and see what secret data can be stolen – never mind, I’m getting a little carried away here.

Go ahead and compile the requisite files, call-procs.c, C_PI.c, and MPI_08_b.c, by entering the command mpicc [file name.c] -o [file name] –lm in each case. The extension -lm is required when compiling C files containing the math header file <math.h>. Execute the file call-procs, or the file name you created, using the command mpiexec -n 1 call-procs. After the first execution, use a different process number in each subsequent execution. Your run should approximate to the ones depicted following. Note that, because of the multithreaded nature of the program, the number of processes executes in random order, and there is clearly no dependence on one another:

Note, from here onward, the run data was from a Pi2 computer. The Pi3 computer will give a faster run time (its processor clock speed is 1.2 GHz, as compared to the Pi2, which has an overclocked speed of 1 GHz).

Execute the serial Pi code C_PI as depicted following:

Execute the M_PI code MPI_08_b, as depicted following:

The difference in execution time between the four cores on gamma (0m59.076s), and the four cores on Mst0 (16m35.140s). Each core in gamma is running at 4 GHz, while each core in Mst0 is running at 1 GHz. Core clock speed matters.

Preparing slave node

You will now prepare or configure the first slave node of your supercomputer. Switch the HDMI monitor cable from the master Pi to the slave Pi. Check to see whether the SD NOOBS/Raspbian card is inserted in the drive. Label side should face outwards. The drive is spring-loaded, so you must gently insert the card. To remove the card, apply a gentle inward pressure to unlock the card. Insert the slave power cord into a USB power slot adjacent the master Pi, in the rapid charger. Follow the same procedure outlined previously for installing, updating, and upgrading the Raspbian OS on the master Pi, this time naming the slave node Slv1, or any other name you desire.

Use the same password for all the Pi2s or Pi3s in the cluster, as it simplifies the configuration procedure.

At the completion of the update, upgrade, and installs, acquire the IP address of the slave1 Pi, as you will be remotely logging into it. Use the command ifconfig to obtain its IP address. Return to your main PC, and use sftp to transfer the requisite files from your main computer – same as you did for the master Pi2. Compile the transferred .c files. Test or run the codes as discussed earlier. You are now ready to configure the master Pi and slave Pi for communication between themselves and the main PC. We now discuss configuring the static IP address for the Pis, and the switch they are connected to.

Summary

In this article we have learned how to prepare master node and slave node, and also how to transfer the code for supercomputer.

Resources for Article:

Further resources on this subject:

Packt

LEAVE A REPLY

Please enter your comment!
Please enter your name here