Massively Parallel Parallella

Parallella with 18 cores: 2 ARM + 16 RISC is the first technical conference for Parallella, to be held May 30, 2015 in Tokyo, Japan.

You can participate even if you’re not in Tokyo. Parallella owners around the world will connect their boards via the Internet to create the most efficient supercomputer on the planet. has built a special micro-SD-card image (using Docker) so adding your Parallella to the mix is relatively quick.

This article includes detailed steps plus a few gotchas.

First: Why?

Some problems are too thorny for a typical computer to solve in a reasonable amount of time. Supercomputers might solve these problems quicker, but they cost millions of dollars. Solution: Link a bunch of $100, 18-core Parallella computers together via the Internet to amplify our compute power while saving money. This shared computing technique has worked in the past:

The benefits are clear. Let’s get started.

Materials Required & Parallella

Here’s what you need to participate:

This procedure is has been tested on Mac OS X Yosemite (10.10.3). Linux instructions are probably similar.

Gotcha: Headless Parallella

Something that tripped me up until I learned more: A Parallella configured for will run headless. There will be no HDMI output. When my HDMI monitor showed No Signal, I thought that I had broken something. No worries. This is just a headless system.

I don’t know if that will change.

Download the Image

Download the image from the site. has created a special SD-card image (using Docker) for this project. As you can see below, I stored mine in a directory called ~/Downloads/Parallella$.

~/Downloads/Parallella$ ls -al
total 2678784
drwxr-xr-x   3 rth  staff         102 May 26 10:53 .
drwx---r-x+ 86 rth  staff        2924 May 26 11:00 ..
-rw-r-----@  1 rth  staff  1371537408 May 26 10:56 resin-supercomputer-0.1.0-0.0.14-Z7010-16.img


Prep for Burning the SD Card

Micro SD card and adapter.

Burning the SD card took about sixteen minutes (once I got the procedure right, as described below). Suggestion: Start the burn process, make sure it’s progressing, and then go work on something else. A watched pot never boils.

Insert your SD card into your Mac’s SD card reader, and use the Mac OS X diskutil list command to determine the designation of the SD card. If you use portable hard drives with your primary machine, the SD card designation could change from time to time, so it’s important to perform this step each time you burn an SD card.

~/Downloads/Parallella$ diskutil list
   #:                       TYPE NAME                    SIZE       IDENTIFIER
   0:      GUID_partition_scheme                        *960.2 GB   disk0
   1:                        EFI EFI                     209.7 MB   disk0s1
   2:                  Apple_HFS MacSSD                  959.3 GB   disk0s2
   3:                 Apple_Boot Recovery HD             650.0 MB   disk0s3
   #:                       TYPE NAME                    SIZE       IDENTIFIER
   0:     FDisk_partition_scheme                        *8.0 GB     disk2
   1:               Windows_NTFS Parallella              8.0 GB     disk2s1


From this diskutil report, we can see that we want to burn the SD image to /dev/disk2. The other device is the hard drive for my primary machine. Burning the wrong device will destroy data.

Burning the Card, Skipping the Gotcha

To burn the SD card…

  1. Change into the directory where you downloaded the image.
  2. Unmount the destination partition on the SD card.
  3. Use the the dd command to copy the image to the SD card.

Gotcha alert: The first time I attempted to burn the image, I used a block size of 64k (bs=64k) for the dd operation. Big mistake on my part because the write operation took almost four hours. On my second attempt, I used a 1 megabyte block size (bs=1m) and the write operation finished in a speedy 16 minutes. So, learn from my pain and use bs=1m.

Burn your micro-SD card like so:

$ cd [directory containing the image file]

$ diskutil unmountDisk /dev/disk2

$ sudo dd if=resin-supercomputer-0.1.0-0.0.14-Z7010-16.img of=/dev/disk2 bs=1m
1308+0 records in
1308+0 records out
1371537408 bytes transferred in 960.001945 secs (1428682 bytes/sec)


Burn progress.

If you want to make sure that dd is running properly, use OS X’s Activity Monitor utility to monitor burn progress (see the ‘burn progress’ diagram). Here’s a quick run-through of the dd command options:

About the New Image

Image files.

The new image will be called boot-parall as shown in the diagram. Depending on your particular image, your list of files may be the same as mine.

Boot Parallella with the Image

Once the micro-SD card has been burned, eject it from the Mac and insert it into the Parallella. Then apply power to the Parallella.

Working Headless

As mentioned above, a Parallella configured this way will run headless, without HDMI output. So, how do you know if the image is working?

First, get a network scanner and use it to locate your Parallella on your local area network. I use an iPad app called Fing, but any good network scanner will do. Your Parallella will appear as parallella-hdmi-resin, along with its corresponding IP and MAC addresses.

Ping the IP address given by your network scanning tool like so:

~$ ping
PING ( 56 data bytes
64 bytes from icmp_seq=0 ttl=64 time=4.767 ms
64 bytes from icmp_seq=1 ttl=64 time=3.694 ms
64 bytes from icmp_seq=2 ttl=64 time=4.057 ms
64 bytes from icmp_seq=3 ttl=64 time=3.969 ms
--- ping statistics ---
4 packets transmitted, 4 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 3.694/4.122/4.767/0.396 ms


If your Parallella is up, running, and attached to your network, your output will look similar to what you see above. But if ping can’t find your Parallella, you’ll get…

~$ ping
PING ( 56 data bytes
Request timeout for icmp_seq 0
Request timeout for icmp_seq 1
Request timeout for icmp_seq 2
--- ping statistics ---
4 packets transmitted, 0 packets received, 100.0% packet loss


Connect to

After you have successfully connected your Parallella to your LAN and the Internet, visit from a computer other than your Parallella (because the Parallella is headless). This page will tell you when your Parallella has connected to the mothership.

To do this, you will need the media access control (MAC) address (also known as the hardware address) for your Parallella’s network interface. I used my network scanning tool, Fing, to find the MAC address for the Parallella. You should be able to do the same with your network scanning tool.

Enter your Parallella’s MAC address in the field on the form, and click search to determine whether your device has connected. searching for your Parallella.

Your first connection attempt might not be successful. It takes awhile for the first connection to happen because the Parallella might need to download software updates first.

It took about 30 minutes for to recognize my Parallella. The successful connection was cause for celebration:

Parallella connected w/MAC address and city. Celebrate!

What’s Next?

I’m not sure what’s next for and Parallella. I would imagine that Status = idle will be replaced by something more interesting, and images processed will eventually be greater than zero.

Engineers at Parallella and have done a great job so far. Now, since it’s likely that they enjoy ideas from the peanut gallery, here are a few suggestions for them:

Hats off to the Parallella and teams. Looking forward to more!