Beowulf Cluster: Part 2: Not a Beowulf Cluster Yet

Good Morning from my Robotics Lab! This is Shadow_8472 and today, I learned a lot toward the next step in my project and backed out at the last minute. Let’s get started!

You know the thing about clusters: it takes some overhead to run them? The computers I’m working with in my Third Workshop care package don ‘t have much CPU to begin with. Special thanks to kevrocks67 on the EngineerMan Discord server for getting that point through my head and pointing me in the right direction: subnetting.

My understanding of subnetting is lacking to say the least. Perhaps if I back up and just tell the story.

This week saw most of the parts I need for my cluster come in. Shopping for the Pi 4, I compared different sellers and chose one to go along with a separate heat sink case I picked out from a different distributor. I managed to save a little money by assembling my own kit instead of getting one that included some small, aluminum heat sinks.

When the case arrived, it came with some thick thermal pads. Yeah, those won’t be separating once the case is fully installed and operational temperatures melt the pads into place or they are just sticky. I did a dry fit and it doesn’t look like the case will operate correctly without them. In the meantime, I noticed the camera cable slot and the similar-looking display slot will likely be inaccessible once I assemble the case, though the GPIO pins will still be open.

My father picked up a couple outdoor grade 3-way power spliters for power management. The minicomputers I’m working with have bulky transformers right on the plug, and they are such a shape as to make it difficult to creatively arrange more than three on a seven-plug power strip.

The Ethernet cable we picked up was paid for with a little effort. One of my church members was wiring up the church a while back for Cat6, and he has some left on the spool. I stayed topside while my father donned a jumpsuit and finished running the cable to just below the office, where he cut the cable. There was enough to reach the office, but that was it: maybe about 30 feet max: more than double my very rough estimation of what I’d need. A package of Cat6 ends and a crimper will see me learning yet another new skill in the near future.

About the only piece I don’t have picked out yet is a board to mount things on. It doesn’t look like the heat sink case has any good places to mount, though the switch and minicomputers do.

The next logical thing to do is assemble an early prototype! One control node and one worker node should be a good start… if only I could get the pi to boot from an SD.

Lacking the desire to format any drives where a valued drive can possibly be hit in the crossfire, paired with the inability to read SD cards on anything but my laptop, I tried using an app on my tablet to flash a card. For one reason or another, I wasn’t able to boot. I suspect I need root privileges to properly format anything with Android, unless I want to format as internal or removable storage. Lacking access to an old phone nobody has no qualms with me “voiding the warranty,” a multi-SD reader is on order.

Speaking of SD readers, turns out each of the five little helpers has one built in. I’ll be using one of those along with a live USB for formatting fun in the near future with no chance of fireworks at the end, just as soon as I figure out why they look exactly the same with lsblk -l. I suspect it’s got its own SD card like storage device in there. Ironically enough, I suppose the difference should make a slip up of the same calibur nearly impossible this go around. Nevertheless, I still want to practice isolation of formatting whenever possible. In either case, what I am trying to think of as equivalent to a hard drive is not labeled sdx anymore.

Somewhere, I made the switch in short-term project aim from cluster to subnet to reclaim some of that overhead. I still want to end up with a cluster for the experience, but for now my token effort among token efforts will be more efficient with less overhead.

To help with overhead costs, I found a site that compares lightweight distros. Thishosting.rocks/best-lightweight-linux-distros/ They have a sortable chart of about 30 distros and their rough minimum system requirements, including RAM, CPU, and space on disk. Folding@Home is likely to use CPU above anything else, I suppose, and RAM can’t be easily expanded, so my dump stat is disk space.

I optimized for CPU, and found Debian was a near-perfect choice. Most, if not all, my Linux experience is in Debian or a decedent of Debian. The required RAM is one of the lower rankings while still needing an almost non-existent CPU. I even already have the install media plugged into the thing and booted! Though I may want to look into a server installation option.

In terms of mentally tangible progress toward a prototype, I set the Pi 4 aside for the moment and brought in BlinkyPie, my Pi 3B+. My goal: turn a Raspberry Pi into a router. I’ll know everything works when I have Internet access on the minicomputer that is otherwise not connected through any other means.

I viewed a few networking videos that were either dated, covering basic material I already knew, had bad audio quality, or showed what to do without giving any theory behind their actions. I eventually settled on a clear, step by step tutorial on at Linuxhint.com/raspberry_pi_wired_router that does not explain its actions, but the steps are short enough that I can seek out my own explanations as I go.

The tutorial starts with installing Raspian and getting SSH going. As anyone who’s been following my blog long term knows, SSH was a real pain for me to figure out, and if I didn’t know what I was doing, it might as well be a small element of a magic ritual to start over if things fail. Since I intended to work with a Pi that was already established, I glazed over those steps and started in on configuring the network.

The whole reason this project of a subnet is possible with the hardware I have is because the Pi can participate in two networks at once: one via Ethernet, and one via WiFi. To make a long story short: this procedure requires the Pi establish itself in a special IP address for the router, implying a static IP. I sometimes take this Pi elsewhere (or at least its chip), and I don’t want to forget and accidentally mess up someone’s network, so I backed out in favor of waiting on a properly running Pi 4.

As always, this project is evolving as I work on it. The Gigabit Ethernet speeds I thought I would use for a cluster won’t be necessary for the time being, as there’s a slightly lower bottleneck at the Pi’s WiFi network adapter. Next week should be interesting.

Final Question: When was your last project where you lowered the specs and found yourself overbuilt?

Leave a Reply