I’ve been sitting on this topic since last summer, which was the first one during which I’ve had temperature measurements and history available for my living room. Said living room is also where my Homelab lives, right next to my desk. I’ve never really minded that, but it sitting right next to me most of the time was one factor in deciding to go with low-power Raspberry Pis instead of a couple of Enterprise grade servers. But now I think I discovered a real problem: Heat.

We’ve had a couple of mild days after the recent heat wave in Germany. And somehow I wasn’t able to cool down my living room properly. The temps go down while I have my balcony door open, but they also go up rapidly again once I close it.

Here is the main figure I want to talk about:

Sorry to the screenreader users. The detailed behavior of the curves is actually important for this blog post, and I know it got a bit wordy. I hope it is still somewhat clear what's going on in the graph. A screenshot of a Grafana time series. On the Y axis, it shows the dates from August 18th 00:00 to August 20th 20:00. It shows two temperature curves. The consistently higher one, shown in orange, is referenced in further text as the indoor/living room temperature. The consistently lower one shows the outdoor temperature. The outdoor temperature starts out at 22.5° C. It stays above 20° C until 18:00 on the 18th. Then it starts falling towards the lowest point at 15.2° at 06:00 on the 19th. It rises a bit throughout that day, but its maximum on the 19th is 18.3° C. It only rises above 20° C in a sudden spike on the next day, August 20th, shortly past 18:00 before falling back to 19°. The second curve, the indoor temperature, starts at 27.9° C. Around 01:00 on the 18th, it starts falling slowly to 26.9° at 08:40. Starting there, it falls more steeply, towards 25°. At around 15:00 it starts rising back to a max of 27°. Then it falls rapidly again to 25° at 20:30. Then it rises back to a max of 27° at 01:00, where the slow fall happens again until 08:40, where the more rapid fall happens again. Shortly thereafter it rises again rapidly, just to fall rapidly again. At 12:15 on the 19th, it rises again, first steeply, then more gradually, to a max of 26.5° at 01:00. Then the slow fall happens again until 08:40 on the 20th, finally followed by another steep rise plus slower rise as before. But the slower rise now peters out at about 26° C around 21:00 on the 20th.

Temperature measurements from August the 18th 00:00 to August 20th 22:00. The orange upper curve is the indoor temperature for my living room with the Homelab and my desktop machine in it, and the red, lower curve is the outside temperature measured on my balcony.

The above temperature measurements have been made in my living room on a wall which is not near either my Homelab nor my desktop machine. The outside measurement was made on my balcony, from under a table. The large spike in the outdoor temperature can be ignored, that’s an artifact from the placement of the sensor and that was probably the sun coming out for a while. The measurements are made with the same type of sensor, but those sensors are not anything fancy. But in this post, I’d like to discuss the temperature behavior, so the correctness of the absolute values is not too important, as long as the sensors are off by a consistent number, which I believe they are.

As I’ve noted above, these measurements were taken between the 18th and 20th of August, which were relatively cloudy days with relatively low temperatures for mid-August. These three days were a welcome respite from the heat, and I wanted to use them to get the heat out of my apartment. Note that I don’t have A/C in here, so any cooling has to come from cool air being let inside the place. This generally happens in two ways in my living room:

  1. During the night, before going to sleep, I half-open my balcony door. I’m not sure whether that’s even a thing in other countries? Anyway, it basically means that instead of opening the door to either side, I instead tilt it just a bit towards the inside.
  2. Once I get up, I generally open the door fully

Before going into the graph, let’s describe the “me” factor in the measurements. I started a one-week leave on Monday, so I was home most of the day during Sunday and Monday, the 18th and 19th. On the 20th, I started a trip to visit family around 12:00 and haven’t been back since then.

When it comes to computers in that room, it holds both my Homelab and my desktop machine, plus two 22" screens. The Homelab, meaning 13 Raspberry Pi 4, five low to mid power x86 machines and a passively cooled 16 port switch eats about 200 - 220 Watts constantly. My desktop uses about 100W - 120W during normal web surfing/Youtube/Dev usage and about 230W when gaming. Plus another 30 or so Watts for the two monitors. I switch the desktop on in the morning and switch it off before going to bed.

Then there’s also the human factor in this, because I was in the same room most of the time, minus the nights, but I will ignore that for the sake of this discussion.

So the base load of the room is a constant ~220W from the Homelab, plus another 150W - 260W from my desktop setup during waking hours.

Now onto a bit of behavioral description when it comes to ventilation. I went to bed every night around 01:00, at which time I tilt my living room balcony door and open my bathroom door and tilt its window as well for some through-ventilation. These account for the relatively slow nighttime drops of the temps throughout the graph. Then when getting up around 08:40, I open the balcony door fully. This event can also be seen very clearly in the graph as a sharp temp drop.

And here we come to the point of this whole post: Once I close the balcony door, the temp immediately rises sharply and reaches almost to the previous max again. And just for those who are wondering: Yes, I’ve checked multiple times whether I accidentally put on the heating.

Note also the difference in the climb from 12:00 on the 19th and 12:00 on the 20th. On the 19th, it climbs quite high, to 26.5° C or so, while on the 20th it peters out around 26° already.

The explanation seems pretty obvious, of course - at the peak, I’m dumping about 480 Watts worth of heat into that room, plus whatever my body produces. I’m just a bit surprised at how rapidly the temps rise after closing the balcony door.

I would have really loved to put some math into this, but my google-fu is utterly deserting me. I’ve gone up to page eight now and not found anything better than “yes, every Watt of electricity consumed is turned into heat”.

This is another reason to get a move on with the Kubernetes migration, as finishing that would allow me to get the Homelab back to around 150W.

In conclusion: I should really be looking for an apartment with a well-ventilated separate room for the Homelab and switch my desktop to rack-mounted while I’m at it.