1

What causes servers to shut down or reload in case of high temperature? For example Juniper network equipment running Junos OS will halt the routing engine if it's CPU temperature remains above 100C for more than 60s:

Jun  3 00:40:32  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_WARNING: Routing Engine 0 temperature (101 C) over 100 degrees C, platform will shutdown in 15 seconds if condition persists
Jun  3 00:40:32  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_WARNING: Routing Engine 0 temperature (101 C) over 100 degrees C, platform will shutdown in 15 seconds if condition persists
Jun  3 00:40:37  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_WARNING: Routing Engine 0 temperature (102 C) over 100 degrees C, platform will shutdown in 10 seconds if condition persists
Jun  3 00:40:37  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_WARNING: Routing Engine 0 temperature (101 C) over 100 degrees C, platform will shutdown in 10 seconds if condition persists
Jun  3 00:40:42  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_WARNING: Routing Engine 0 temperature (102 C) over 100 degrees C, platform will shutdown in 5 seconds if condition persists
Jun  3 00:40:42  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_WARNING: Routing Engine 0 temperature (101 C) over 100 degrees C, platform will shutdown in 5 seconds if condition persists
Jun  3 00:40:47  M10i chassisd[1209]: CHASSISD_RE_OVER_TEMP_SHUTDOWN: Routing Engine 0 temperature above 100 degrees C for too long; powering down all FRUs

However, what about servers? Does this depend on server hardware vendor? Or does this depend solely on server operating system?

Martin
  • 352
  • 3
  • 12
  • 29
  • 2
    Q: What causes servers to shut down or reload in case of high temperature? A: High temperature. - Do you mean to ask what mechanism in the server will initiate a shutdown when a temperature threshold is exceeded? – joeqwerty Aug 01 '13 at 14:59
  • @joeqwerty exactly. In addition, what determines those thresholds. – Martin Aug 02 '13 at 08:04

2 Answers2

5

Does this depend on server hardware vendor? Or does this depend solely on server operating system?

Either/both - if a server has temperature sensors, and some don't, then they can have hard limits set in BIOS usually, but the OS or drivers inside could override those thresholds for whatever reason.

So yes, either/both - you need sensors either way, and this is one reason I buy HP :)

Chopper3
  • 101,299
  • 9
  • 108
  • 239
  • 1
    This means that if for example server CPU temperature is about to reach the critical BIOS threshold because environmental conditions in COLO, one could override this BIOS threshold from OS if he is sure, that this does not damage the server hardware? How to do this under GNU/Linux? – Martin Aug 02 '13 at 08:34
  • It's entirely dependent on the system and its drivers - there's no one-size-fits-all I'm afraid – Chopper3 Aug 02 '13 at 08:35
  • ok, I see. However, is it a common practice to monitor temperature sensors in OS using for example utilities in `lm-sensors` and write scripts which initiate an action(for example shutdown) if certain threshold is reached? I guess it's not widespread and most rely on BIOS/UEFI thermal thresholds and actions? – Martin Aug 02 '13 at 08:52
  • Can really speak for everyone but we use the HP driver stack (called CIM agents) and they automatically add system-wide alerting for their temp sensors into all common Linux's and Windows. – Chopper3 Aug 02 '13 at 09:29
  • Most(may actually be all but I am not familiar with every single processor architecture) Intel CPUs, which make up most of the CPUs in servers, have on-die temperature sensors as well as power consumption monitors and controls. While there is typically some access to these settings in BIOS, others like high, critical and max temperature are hardcoded on the CPU. These sensors also link to shutdown and throttling mechanisms that are beyond user or OS control once passing certain values. While there may not be a one size fits all answer, for most server CPUs this will be the case. – Matt Oct 15 '16 at 16:51
3

It's a mixture of both, but you need temperature sensors in place to do anything. So... make sure you get equipment that has it.

Most hardware based solutions have a thermal trip threshold in the BIOS itself. Depending on OS support, a thermal trip can look like a sudden power failure (never mind all that fancy power protection you have) or as a user-initiated safe shutdown. In the case of embedded devices like network switches the line between hardware and OS is fuzzier; your JunOS device is smart enough to turn off the high-voltage components while temp is above a certain threshold.

sysadmin1138
  • 133,124
  • 18
  • 176
  • 300
  • Is Linux kernel (usually) able to initiate a safe shutdown if thermal threshold of BIOS is reached? – Martin Aug 02 '13 at 08:47
  • @Martin To my knowledge, not natively; it requires services/drivers from the hardware manufacturer to set up. "OS Support" in this answer includes additions from OEMs. – sysadmin1138 Aug 02 '13 at 11:24