As posted last year, we decided to use a low-power consumption machine as the very first QuadraSpace server. The goal was to embrace the green computing philosophy, that is trying to limit the energy needed to run the IT infrastructure of the project. Low-power consumption on the server means also low-power consumption on cooling systems and, at the end, low-noise pollution.
Since last January, we were experiencing strange unexpected power loss on the server. We thought it was caused by a recent kernel upgrade of the Ubuntu Server 9.04 distribution. Some posts in dedicated forums stated that there could be issues with the ACPI support on ION-based boards (ION is NVIDIA chipset that runs on the Acer Aspire Revo R3600).
After weeks of investigation, we found that the issue was caused by the thermal protection of the CPU. In certain condition (but we were not able to determine what was the actual reason), the operating system powered off the server to protect the hardware. However, there was not relation between CPU load and unexpected power off.
We moved from software to hardware. Apparently there were no issue with the CPU fan. Apparently. We powered off the server for about one hour, just to replace some cables, then the server was powered up again. After a while, the was a strange noise coming from the server: some “crick”, “crack” on the plastic covers. In less than a quarter the server case was very hot and the fan was not working. After a while, the fan started to rotate very slowly, but the board was so hot that Ubuntu powered off the system.
We opened the case and we discovered that the fan was not working properly.
The REVO opened
After a power cycle, the fan may rotate very slowly or even be completely blocked. However, at the very start of the system (where the fan is powered-on for a bit, without any speed control), it always rotated at maximum speed. We deduced that there was an issue with the fan speed control.
The fan removed from the heatsink
Therefore, we decided to completely exclude the fan control (which is performed using a PWM signal on the blue cable) and to leave it to run at maximum speed.
The BLUE cable has been cut: no power speed control will stop the fan!
The QuadraSpace server is now much more noisy, but no more unexpected power-off have been experienced.
Keep sensing, now…