Related video: QILIPSU Outdoor Enclosure with a Computer Inside… Because.
Visit the Data Grave coffins: outdoor.s.co.tt
Hi, I’m Scott and today we’re going to talk about a couple of computers I buried in my backyard to create a data graveyard.
They’re Raspberry Pies, which are great for this purpose as they’re compact and consume very little power, meaning they can be supplied by power over ethernet and won’t cause rampant heat dissipation issues.
But they’re also pretty good for their intended purpose: Backups.
If you’ve seen a couple of my other videos, you know I tend to go on rants about backing up data. For me, and many of you, most of the content I generate is digital. Losing all my data would be tantamount to someone’s house burning down in the pre-digital age. All my stuff is in there.
And in fact, house fire, floods, natural disasters, burglaries, war, seizure and all sorts of other catastrophes can lead to the destruction of your data. Which is why I always advocate for off-site backups, so your digital possessions aren’t tied to your physical ones.
So in that way I was considering alternatives to traditional off-site backups, and I came up with the idea of the Data Grave.
It’s kinda tongue-and-cheek / kinda serious. Do I think burying your data in your yard is the wave of the future for data security? No, of course not. But I do think it’s almost a viable secondary or tertiary backup strategy.
You might consider it your only offsite backup solution if you don’t want to store your data with third-party companies like Google, Apple, Backblaze, and so forth, and if you don’t have a secure alternate location in which to situate a server that’s 100% under your control.
With a deep enough burial, your underground data is likely to survive a nuclear apocalypse… even if you probably won’t.
I should say, this video is a proof-of-concept. The way in which I prepared and entombed the two computers in their coffins isn’t necessarily optimal. I’m just testing it out at this point, and I plan on revisiting the project in a year to see how it’s going.
Actually, since you might be watching this video quite a while after I uploaded it, you can keep an eye on how it’s going in real time. Over a year ago, I mounted a full computer in an outdoor enclosure to see how it would hold up (link in the video description), and for this project I changed out the computer but left the enclosure and the website up. Go to outdoor.s.co.tt, and that site is served by the computer mounted to the side of my house. At the bottom of the page are links to each coffin. The web pages are exceedingly simple, but they’re hosted on web servers in each coffin. If the web pages come up, then the underground computers are alive.
Technically, the concept is pretty simple: Use a Raspberry Pi (or other low-powered single-board computer) as a backup-slash-storage server. Put a massive SD card in it, and maybe even attach some USB drives. Those drives could (and should) even be put into a software RAID array. Then bury the whole thing in your back yard.. or wherever.
So, if you’re interested here’s how it all came together.
I started with a couple of Raspberry Pi 3 B plusses. They’ve got 4-core ARM Cortex SoC’s running at 1.4Ghz, 1GB RAM, four USB 2.0 ports, and an ethernet port that’s going to become extremely relevant. They’re neither fast nor powerful as computers go, but are more than sufficient for use as a home backup server.
Their low power consumption and ethernet port are extremely important because they can be combined with a power-over-ethernet hat (made by LoveRPi in this instance), meaning they each need only one cable connecting them to the outside world, which will provide both power and data.
It’s called a “hat” because it sits atop the Pi, interfacing with a few select pins for power and ethernet, while not obscuring the CPU or other pin headers. Well, a hat usually obscures your whole head, but in this case it’s a double entendre and HAT stands for “hardware attached [on] top”.
With the hats attached, I imaged a pair of micro SD cards with the latest version of Raspberry Pi OS, a Debian variant. Aside from the hostnames, both installations are pretty much identical from start to finish, but one card has a capacity of 256GB and the other only 32. When using them for backups, I’d probably keep the SD cards small and used USB drives in RAID for the actual data storage.
I chose to install the full GUI, which is a LXDE desktop environment running the Openbox window manager. The GUI is completely unnecessary for a backup server, but I figured in this case it would be good for load testing, as well as being more photogenic when appearing in a video.
I also tested out the PoE hats using a TP-Link TPE-S44 ethernet switch with four standard and four PoE ports. It has a maximum capacity of 15.4 Watts per port, which is more than enough for the Pi. It consumes about 2 Watts at idle, while the maximum consumption depends upon workload and devices that are attached. But in any case, it’s much less than the maximum of this relatively inexpensive switch.
To help keep connectors from corroding, my plan was to coat them all in dielectric grease. This includes the SD cards and slots. Here I used a 3M silicone product which is intended for automobiles, but should be more than suitable in a Raspberry Pi… probably.
Being as these Pies would run in a truly headless configuration, I configured their VNC servers and tested them with only an ethernet cable attached to verify overall functionality. And not for the last time, either, as you’ll see.
When mounting a computer outside in the elements, let alone underground, the main issue is water. I suppose if you live in the desert then maybe it’s less of an issue, but it still does rain on occasion. Given enough time, water will get into pretty much anything, even a decent quality supposedly “water tight” case. But an allegedly impervious container is as good a place as any to start, for mechanical protection at least.
So I decided to use a couple of small Pelican cases. Both are relatively inexpensive, and both are large enough to fit a Pi with USB drives attached.
But the cases aren’t going to be the only differences between these two.. coffins. In the yellow one, the Pi will sit in there with only an ethernet cable attached. In the grey one, there will be dongles to allow for debugging in the future, should the networking fail at a hardware or software level. I used very short USB and HDMI extensions so a display and input devices could be connected. In retrospect, extending the mini USB power input port may have been wise, as a failure of the network port or PoE hat could render the device powerless. Though, I wonder if you can backfeed 5 volts into the Pi via the USB-A host ports? Perhaps one day I’ll have to find out.
The ethernet port was also extended, but with a theoretically water-tight connector that has an RJ-45 jack on the inside.
Of course, those extender dongles aren’t necessary, unless the actual ports on the Pi are somehow inaccessible. And indeed that’s what will happen, because the next step is to pot the Pies in with epoxy.
But first I prepared a couple of USB flash drives by removing their cases. The idea being that any void spaces sealed in the epoxy (like the insides of the plastic shells) will contain air that will inevitably have some moisture content. I keep the humidity low in my basement, but at a sufficiently low temperature some water might condense. Plus this will let the flash chips bond directly to the epoxy, helping to wick heat away.
I used two different types of potting compound. The yellow coffin will use a clear compound to allow us to see the Pi as it sits in stasis, LEDs blinking away. More importantly it might help with fault analysis without the need to remove the solidified epoxy, when the device inevitably fails. (Though hopefully later rather than sooner.)
The grey coffin will receive a black, extremely opaque formulation. But that formulation has the benefit of being thermally conductive, which should (and as we’ll find out, indeed does) keep the Pi running cooler. The downside is that it’s going to be quite the forensic archeology project to reveal any physical faults. But I couldn’t find a compound that was both clear and thermally conductive. (Not saying it doesn’t exist, just that this is what I ended up buying.)
I don’t know how necessary this was (and it probably wasn’t), but I filled all of the unused connectors with more dielectric grease prior to potting. Realistically, if water infiltrates the potting compound then it’ll effect all parts of the PCB, not just the connectors.
The connectors that were used also got greased up, but those I justify by the thought that water might wick through the cables into the connectors specifically. Hopefully the grease will help to ameliorate that problem.
To save some space and reduce the amount of potting compound needed, I filled some unused volume in the grey coffin with open cell foam. Because the foam would absorb the epoxy, I covered it in electrical tape (the best kind of tape) before pouring.
If you ask me now which of the two concoctions was better to work with, I’d say the clear compound without a doubt. I’m not sure if it was due to the thermally conductive mix or some other difference, but the black epoxy was ridiculously viscous to the point that it was difficult to stir and ultimately pour. It also smelled horrible.
But time will tell if it’s the better choice, as it may hold up better in its earthen environs.
In any case, it wasn’t untenable, and I successfully mixed up my first small batch. The directions said to let it stand for 15 minutes after mixing to de-air. I’d actually purchased a cheap vacuum pump and vessel to remove bubbles in a faster and more thorough manner, but due to personal reasons (the pump being loud as shit and my wife being asleep) I decided to de-air it au natural.
With that first batch, I coated the bottom of the grey case with a few millimeters of the stuff. Then with the remnants, I thoroughly coated the back of the Pi. I think this is a necessary step to ensure that the PCB gets full coverage and adhesion on the underside, as just plonking it into a pool of the stuff might leave bubbles against its underside.
I then pressed it down a bit and brought it up slightly, so that the board wasn’t in contact with the bottom of the case. I had considered using stand-offs to keep an adequate layer of potting material between the inside of the case and the bottom of the PCB, but that would have created an interruption in pottedness at those points so I opted to finesse the board a few millimeters above the bottom of the case and leave it at that.
The potting compound was so thick and tarry that I don’t believe the Pi was able to sink into it.
Then I created a batch using the remaining compound, which was quite a lot. That harkened back to the part of the instructions that said the 2 hour working life specified was based upon a batch size of 100 grams, and that the working time would diminish in inverse proportion to the batch size. I measured it at about 475 grams (minus the cup) and — worried about de-airing it for too long — I poured it pretty soon after. I figured it could de-air in the coffin, and the bubbles would rise away from the PCB anyhow.
Then it was the turn of the yellow coffin, and I probably needlessly injected dielectric grease into all of its unused connectors. The potting compound probably would have flowed in and filled them completely regardless, but I felt that it was the best move in case moist-ish air did get trapped within them.
The yellow coffin – let’s call it the “yoffin”.. wait, no, that’s awful – also wouldn’t have any of the fancy dongle trappings of the grey coffin (Goffin?). It would just be a straight run of CAT 6 through a hole in the case and then into the Pi.
I measured the cable and made the hole as tight a fit as possible, but of course that’s not gonna stop water from infiltrating around it. But the hole will be below the level of the potting compound, and besides, I stripped back far more of the cable sheath than necessary. This way the individual conductors will become encapsulated, providing a break for any water wicking inside or around the cable.
The Pi also got one final boot-up before potting, just as did the one in the Goffin. Everything checked out, so it was time to pot.
The clear fluid, being less viscous, de-aired quite effectively just sitting on the table. It was a mix of about 375 milliliters total, which was the entire contents of the two containers poured simultaneously.
I poured it over the back of the PCB first to ensure coverage, and then flipped it over and continued pouring to get it completely covered. That 375 mL was more than enough to fill the Pelican 1040 case. Incidentally, the hard plastic exterior of the case is made entirely of clear plastic. The yellow insert is a flexible rubber-like material which to me didn’t seem desirable as it would thermally insulate the potting compound from the earth, to which the epoxy probably wouldn’t adhere all that well. But the lip of the yellow insert makes up the seal for the box, so without it water would just be able to flow right in. In retrospect I would have carefully cut off that lip, used it as a seal, and disposed of the rest of it. That would have also allowed us to view the underside of the PCB during failure analysis.
It was also a little tricky getting the CAT 6 conductors bent into a good way to keep the board neatly positioned. But in the end, the Pi was fully submerged, and the cable sheath was far away from it.
It was then time to cure the epoxy, and initially I was just going to let it sit out at room temperature which could have taken a maximum of 96 hours. My hope was that the relatively large pours would have resulted in faster curing, but after about 24 hours both mixes were still soft.
The instructions mentioned heat curing, but at relatively low temperatures. The lowest my oven would go was 250 degrees Fahrenheit, but I got impatient and decided to use the oven door to regulate the temperature. A few comically overblown multimeters with really horrible thermocouples were deployed to measure the temperature of the coffins at various positions. (The idea being that the multimeters could graph the temperature, but that was both not helpful and the displays too small to monitor while standing near the camera.)
After roughly half an hour each at anywhere between 100 and 175 degrees Fahrenheit, the potting compound seemed to have solidified – on the surface, anyway.
If you saw my previous video on the Qilipsu outdoor enclosure, you’ll know what this box is. It spent a little over 15 months attached to the outside of my house, braving a freezing New York winter and a couple of hot and occasionally stormy summers to test the enclosure.
Inside is an old Celeron mini-ITX motherboard which was incredibly shitty and slow, but which hosted a web server that announced to the world whether or not the system was up and running. Happily it never went down due to any kind of failure of the computer nor the enclosure, but I figured it was time to resurrect the project in a new and improved form.
The case received a new single board computer in an industrial style chassis – the branding on it is “V-N-O-P-N” (VN Open?) – which has a newer Celeron J4125 quad-core CPU, 8GB RAM, a 128GB mSATA SSD, as well as onboard WLAN and four ethernet ports.
That was paired with a TRENDnet TPE-S44 8-port ethernet switch, 4 of which are PoE-cabable.
The switch uses a 48V power supply that match the PoE voltage, whereas the computer uses a standard 12V PSU. I would have preferred one supply split on the DC side to both devices, but I instead mounted both of them in the enclosure.
The components were all attached to the backing plate of the case using industrial strength Velcro – which has a really sturdy adhesive backing and strong hook and loop connections – but the real holding is being done by zip ties.
The holes in the Qilipsu backing plate are only large enough to pass very small and weak ties, so I marked and drilled some of them out for these larger colorful ones.
The purpose of this wall-mounted computer would now be to act as a router to connect the Coffins to the outside world. Ethernet would be passed to the outside using the same connector boots as I installed in the grey coffin.
Those connectors have ethernet jacks concealed within, and cleverly the cinch nut is large enough to fit an RJ-45 connector, and the bushing is split to go over a cable. It means you can pre-terminate cables before connecting them, which is really the main advantage of using a connector like this rather than running the cable through the side of the case and then terminating it, like I did with the yellow coffin.
To distribute power to the two PSUs, rather than put a power strip or multi-tap NEMA socket in the box, I used a Y-splitter cable that I had lying around, in addition to an IEC C-14 to NEMA 5-15 adapter cable so that I didn’t have to replace the existing inlet power cord.
The mini-computer booted up just fine, and I set about configuring it as a web server and router.
If you’re watching this years from now, I may have taken this project offline. But you can go to outdoor.s.co.tt to see if this system is still up and running.
Before burying the coffins, I tested out the whole Data Grave setup on the bench one last time. I mean, if it didn’t work it was a little late to fix any hardware issues with the potted pies, but aside from a minor DHCP issue the whole thing came together without a hitch.
(The Pies use DHCP to obtain their IP addresses, making it easy if I need to connect them to a different net for debugging. The router has static DHCPD entries based upon MAC addresses to ensure that the coffins’ IPs remain consistent between reboots.)
The entombment had far less fanfare than a normal burial.
For this test setup, I located the hole right next to my house. If one were to do this for real as a serious backup solution, I’d dig the hole farther away from anything flammable or destructible.
Obviously there’s grass here that I didn’t want to ruin, so I flayed the top soil as you’d do when cutting sod. Only without a real sod cutter, so it was a bit more awkward than it could have been, but the result wasn’t bad.
Oh yeah, and when you’re digging a hole in this sort of circumstance (to bury electronics in?) it’s a good idea to put a tarp down to throw the fill onto. Makes cleanup much easier, and doesn’t ruin the grass.
I only dug down about two feet, which is above the frost line here in New York – the depth to which the ground is likely to freeze in winter. That means the coffins will probably be encased in icy soil for some of the winter, and they and their cables will be subject to strain as the ground heaves and settles. So it’s probably wiser to look up the frost line depth for your area (if the ground even freezes where you are) and bury deeper than that. Here, that’s anywhere from 30 to 50 inches, depending upon which website I’d want to believe.
In my defense, my back is shit and I hit the much harder sandy subsoil, so I gave up at two feet. But hopefully the subsoil should drain reasonably well and if it doesn’t we’ll all find out together when I pull waterlogged Raspberry Pies out of the ground that I put there for no reason.
When you’re burying your computers, it’s probably a good idea to put a layer of stone underneath them to facilitate drainage. The stone should be surrounded by landscape fabric to prevent earth from infiltrating and filling in the spaces between the stones. Here I used landscape fabric, but it probably wasn’t super necessary. I figured it would at least provide some protection to the coffins.
There was one last step to preparing the grey coffin. Because the yellow one had its ethernet cable subsumed by the epoxy, I had to cut a very long tail for it of about 50 ft which was way too long for safety. To make more efficient use of the somewhat expensive outdoor/direct burial CAT6, for the grey guy I ran the tail from the spool directly into it so that I could measure it out accurately by laying the cable.
The cable and grommet got plenty of silicone grease, as did the unused connectors inside the case.
Then I sealed them shut using a couple of plastic and one metal zip tie each. This was to prevent the latches from popping open either during the burial process, or from the stress of the freeze/thaw cycles.
With the Qilipsu re-attached to the wall and powered up, all that remained was the funeral.
I put a loop of extra cable from each coffin underneath them, to keep strain off of the inlets to the boxes during burial and freeze/thaw.
One last test was in order before shoveling the soil in, so it’s back to the router to terminate and connect the other ends of the ethernet cables. I decided to give the weatherproof jacks a proper test by not putting any dielectric grease on those connectors. Of course, they’re at the bottom of the box so water won’t rampantly flow upwards. But you always have to keep capillary action in mind, where water can indeed work its way into an enclosure against gravity.
Anyhow, the coffins both powered up fine, and so the entombment began as daylight faded.
Uh, apologies for the absolutely crap framing here where I cut off the edge of the hole, but you get the idea.
I tamped down the first layer of dirt by stomping on it to ensure that the coffins were firmly in place. In retrospect I should have stomped another layer because the ground has settled a bit in that spot. But if you’re doing this, then you should leave some loose dirt on top before replacing the piece of sod, but you can definitely compact more than I did.
Speaking of which, the sod was a little unwieldy in that piece as I’d taken a ton of soil with it. So I cut it in half before replacing it. Absolutely no harm done if that happens to you.
Last but not least, I watered the area. This wasn’t to test the water tightness of the coffins or anything, but it’s always advisable when laying sod. Regular watering will obviously keep the grass alive, but also promote root growth into the soil beneath, bonding it all back together.
And such is the story of the Data Grave.
I’ve been pretty busy lately, so I’m writing this script almost 2 weeks afterwards. In the interim there have been a couple of good rainfalls, and temperatures have been creeping down towards freezing. So far both underground Raspberry Pies are fully functional, and responding to requests.
When you go to outdoor.s.co.tt, you’ll see a link to each coffin at the bottom. The web pages are really nothing special to look at, but they are served from their subterranean location.
Being as this is a proof of concept, the story isn’t over. In about a year – or when both Pies fail – I’ll dig the data coffins up to see how they fared. So if you’re watching this video in 2024 or later, check my channel for the conclusion.
At some point sooner than that, I’m going to be posting a video about another coffin that’ll be serviceable rather than potted. So if you’re watching this after the second quarter of 2023, check out that video, too.
