oh, ok.
that was a part of pipe.
input was from this command:
prtpicl -v -c temperature-sensor
I couldn't use that other command for temperature, as it only displays ok/fail in Solaris10.
(hopefully) the output is CPU and server temperature. should be in Celsius?
the plan is to run it in cronjob once per day. so I will see if temperature rises suddenly. it will tell us if one of the CPU fans has failed.
the 3 wire fans are impossible to obtain. also Sun used to sell only heatsink+fans combo, which cost a lot. yes, it is small compared to the price of server, but buying anything is complicated here.
This would require to be tested : unplug a fan and monitor the temperature to see how fast the temperature will increase.
Note that to secure the test, you must benchmark it under high CPU load condition to build the "worse case" possible and see the behaviour of temperature increase, you will then be able to determine what would be a reasonable check frequency to detect when the temperature increase too much.
(I would not be stonished if the temperature of a CPU whose fan is broken and under high load could increase to more than 80� in less than an hour.)
I see.
but could there be any problems running 'prtpicl' every hour? (because it accesses lowlevel hardware)
the server is not under heavy load.
I doubt both fans would fail at same time. and with just one fan, temperature is less than 60deg. (there is a bunch of those larger fans in case that help too)
well after a day or so, there are no problems with the script.
/offtopic
it occurred to me that a simple solution would be to monitor the difference between two temperatures (CPU and case) at same instance of time. because the chassis temperature doesn't change if CPU fan stops working.