Search This Blog

Wednesday, March 21, 2018

Homelab cabinet thermal managment

My homelab's cabinet is a 12U APC NetshelterCX (actually a Kell Systems PSEM, but they're basically the same) soundproofed cabinet. It has a thermal rating of 800W with the stock fans, though my servers running anywhere near that would shutdown due to too high inlet air temperature. Then I sold those and bought a 1600W one.

Thus, I had to build a more powerful fan system for the cabinet. The original bracket was cut and bent sheet steel with a heavy coating of black paint. This screwed into a large slot in the side of the cabinet, which fed into a sound-baffled passage which exited out of the bottom of the cabinet.

This involved milling, bending, and hammering a piece of sheet aluminum, selecting and installing fans, selecting and installing a 24V power supply, and selecting and installing a fan controller. Even though it has large fans, it protrudes into the cabinet less than the original.

Old and anemic

New and awesome

Turning vanes. An obstruction in the cabinet prevents the middle one from having a set.

Laser cut guide vane holders plus cover plate in case I decided to remove the middle fan.
Feature too small...cracked

Sealed with blue RTV
Installed. Hanging wires: I needed the fan controller for another project...

I went with 3x 120x120x38mm high flow rate axial fans I found on ebay for cheap. I added fins to redirect the flow on the underside of the bracket. These were made from laser cut ABS and hand bent super thin sheet aluminum.

I purchased a cheap chinese PWM fan temperature controller from eBay. I taped it and the thermal probe to the inside top of the server cabinet, and it worked...but poorly. The temperature control chip PWM ran at 30 Hz, which was way too low. It was also underdamped in the operating band, which caused the fan speed to oscillate widely. Pretty sure it wasn't a noise issue. I tried everything I could think of, including constant temperature sources, adding thermal mass to the thermocouple, and grounding/shielding wires, but none of it helped. I'd avoid these, though if you do get one, make sure you get the one with all of the fets populated (higher current). The 2A or 3A ones have one missing, but the 4A one has all of them. The one in picture is of a 2A one, but I received a 4A one...classic chinese ebay item.

The eBay temperature controlled PWM fan controller I bought

Brief segway about fan control. There is a lot of confusing terminology out there regarding small dc fan control. 2 pin fans just have power and ground. These can be controlled either by varying voltage linearly, or by PWM'ing the power line. The former only works down to ~half the rated voltage for most fans or they don't have enough power to start. The latter requires a PWM fan controller (like above). 3 pin fans have an extra wire that outputs the tachometer readings. This is useful for measuring fan speed if the power source is constant, i.e. not PWM'd. If the power is PWM'd, then the sensor is, too, which usually messes up its readings. 4 pin fans have power, gnd, tach, and control wires. In addition to the two methods mentioned for two pin fans, these have a third option for control. In stead of PWM'ing the power wire, a low voltage/low current PWM signal can be sent to the control wire. The fan's internal electronics then handle the actual power PWM'ing. This has the added benefit of not screwing up the tach sensor readings because the voltage on the power wire is still consistent. Unfortunately, finding a cheap controller for these fans is difficult. There are some chinese ebay temperature control ones that are meant for 4 pin fans, but I'm skeptical how well they work. Noctura makes a ~$20 manual pot one, but that's the only one I could find. Anyways, back to the fan bracket...

I ended up buying a cheap chinese manual pot PWM fan controller, which works better, though still has a very slight oscillation. The control band of the pot is also very small...maybe the first 3 degrees or so, the rest being full power.

The pot fan controller I bought...supposedly rated for 60V 20A. lol

Turns out that the 3rd fan wasn't doing anything...the flow rate didn't increase when I added it. The 2nd fan didn't add much either. Because the fans are in parallel, and the exhaust passage has a high static pressure drop, I think I'm on a portion of the system pressure vs. flow rate curve that would be better served by high pressure rather than high flow rate fans. I had never designed an exhaust fan system before, but I guess this is why bathroom vent fans are blowers and not axial. I may redo this design to use blowers if it gets too hot in the cabinet.

Update from Nov 2018: I discuss this in more detail in another post, but I tried measuring the passage pressure drop and found it was actually pretty low. I ended up switching to much quieter, same flow spec, lower pressure spec fans and the temperatures seem stable. Still need to figure out automatic control. I currently have to open the back of the cabinet every time to I use the server to plug/unplug the fans, which then operate at full throttle.

Yes, I did all of this for what is essentially an exhaust fan system. I hadn't really made anything in over a year, so I was suffering withdrawal, so don't judge.

No comments:

Post a Comment