TL;DR; The optimal cooling system mode configuration of the Supermicro server does not ensure the stable operation of the LSI MegaRAID 9361-8i controller in a cold data center.
We try not to use hardware RAID controllers, but we have one client who prefers configurations with LSI MegaRAID. Today we faced overheating of the MegaRAID 9361-8i card due to the fact that the platform did not sense the overheating, while the RAID controller did sense.
The view of the platform with the RAID card is shown in the images below:
Several important points related to this proxy server and the operating environment:
The engineer who assembled the platform specifically installed two fans opposite the card, knowing that LSI controllers get very hot. Note that the motherboard barely extends to the controller, ending 3 cm after the PCI-E slot.
As you can see, all fans are connected as standard to the Supermicro motherboard and in the Optimal mode "blow" depending on the sensors on it, the CPU temperature.
This platform has a Xeon E-2236—a quite cool CPU, which presumably did not heat up significantly for the client.
The data center where this server is located is quite cold—the cold aisle provides 18-20 degrees.
The combination of these factors led to a rather interesting phenomenon—overheating of the RAID controller.
The likely chain of events on how this happened
- the cold processor and motherboard informed the fans that they could blow weakly.
- The motherboard under the RAID was absent, and there were no sensors to detect overheating.
- The fans, being configured in the Optimal mode, blew weakly according to the needs of the motherboard and CPU.
- The controller, not receiving sufficient airflow, overheated.
What we did
Switched the fans to "Standard" mode; if necessary, we will switch to a higher performance mode.
Conclusions
Most likely, if the cold aisle of the data center were not so cold, or if the client had been using the CPU intensively, this problem might not have occurred, as the fans would have operated in a more intense mode.
For ourselves, we decided to definitely change the fan operation mode for servers the RAID from Optimal to a mode with increased rotation frequency.
Source: habr.com
