Cluster: thermal testing

I cleaned up some of the wiring and made everything neat in the cabinet.

The next step was to check my modified heat extraction system to see if it could handle the thermal load. The cabinet was originally designed for 800W max, but I'm pushing ~1300W inside the cabinet (~1600W including external desktop). I taped my multimeter's thermocouple to the top, back inside of the cabinet and ran the wires out the wire passage of the back door. I then ran the all-node benchmark openfoam case for ~10 minutes. At full fan speed, the temperature leveled out at ~34-35C, which is good. The more important temperature measurement is the inlet to the server, particularly the node furthest from the air intake slit, which is located on the left side of the cabinet just behind the front door. Node005 is the top right of the 4 node 2U SM server, so it should have the hottest inlet temperature. I re-ran the case with the thermocouple taped to the top of the server near the front right. Temps never got above 31C at full throttle, so that's good. The heat extraction system is adequate.

The only problem is the noise. The fans are way louder than the server hardware in the cabinet, i.e. if I turn off the extraction fans, I can barely hear any noise from the cabinet, but when I turn them on, it's super loud. The fans didn't seem this loud when they weren't mounted, so maybe something is resonating. I'll need to mess with it some.

The first time I ran the extended benchmark case for the thermal tests, I got a segfault in node005. The next time I ran it, it didn't happen. I've run all my memory through memtest (all passed), so I'm not sure what happened. I'll have to watch for segfaults.

So, to do:

1. Fix fan noise problem
2. Fix RAID1 data storage drive
3. Compile guide

Update: I 3D printed a mount for the fan controller that replaces the fan blanking plate in the middle slot (I only used 2 out of 3 fans). I added a provision for the aquarium tube I used to create a water manometer for the Phi testing, which gives me a static port in the fan duct just past the fans. This should give me the static pressure generated by the fan(s), which I can use to calculate the flow rate from the published pressure vs. flow rate curve. For the first test, I removed one of the fans and put the blanking plate on. The bracket doesn't perfectly seal in the air, but it's pretty good. I then connected the fan directly to the 24V power supply and measured the static pressure...the manometer measured 0. Odd. I was certain that the pressure drop was so large that the fans were almost stalling, but apparently that's not the case. I added the second fan and repeated the test. The water rose ~0.25mm, so total pressure differences was maybe 0.5mm. Either the pressure measurement is wrong, or the fans are actually operating at close to full flow rate. I'm not really sure. I'd expect some fairly significant deltaP in the exit duct due to the sound baffles, so I was expecting a fairly high static pressure measurement, but apparently that's not the case. Unfortunately, there aren't a lot of options for significantly quieter, similar flow rate, but lower pressure fans. There are some lower flow rate ones, but I can just throttle these back and achieve similar noise. Not sure where to go from here.

Update 2: I found some lower flow rate ones that are significantly quieter. If I use 3, and assuming that the pressure drop really is that low in the passages, it should have about the same total flow rate as the two loud fans, but be 8 decibels quieter for one option and 14 decibels quieter for another option. Unfortunately, the latter fans are harder to get in the UK (more $$). I can get the former fans from China for fairly cheap. Regardless, this is going to take a few weeks to fix.