Stefaan A Eeckels wrote:
> Hi David,
>
> On Sat, 18 Jun 2005 20:52:30 +0100
> Dave <nospam@nowhere.com> wrote:
>
>
>>I'm wondering if I can use my Sun as a computer controlled heater, to
>>stabilise the temperature! Using as heating elements the 4 CPUs, and
>>adjusting the heat they produce by the amount of work they do, and/or
>>taking them offline. (I bet you have not known of an idiot trying that
>>one!!!)
>
>
> You're like my son, always using things for purposes they were not
> intended for.

Very often that solves problems.

> I'm trying to console myself with the idea he might be
> the guy to solve the world's energy problems.

I think the word "Americans" is appropiate there, but I will comment no
futher.

>>Would taking the CPUs offline make them produce less heat or not? I know
>>when the Sun is doing a lot of CPU intensive stuff, it certainly runs
>>warmer.
>
>
> Yes. CPUs use less power when they idle. To be absolutely sure, use
> prtdiag to see if the temperature of the CPU drops when you take it
> off line (do you get CPU temperature readings on a U80, on a Blade2000
> it shows both the fan speed and the temperature of both CPUs).

I know they use less power when idle, but do they use less when taken
offline with psradm, than when online, but idle?

At the moment its inconvenient to locate my temperature sensor on the
output of the Sun's fans, but I'll get some longer leads on it and try.


>>I'm wondering if I can get the Sun to do more/less work, to help
>>stabilise the temperature.
>
>
> You can parse the prtdiag output for the ambient temperature (which is
> more or less related to the room temperature) and stop or start a few
> CPU-intensive programs or take CPUs off-line as required.
>
> $ prtdiag -v | grep Ambient
> +em-board/cpu0 Ambient 27C -10C 0C 40C 60C okay
> +em-board/cpu1 Ambient 26C -10C 0C 40C 60C okay

No such luck on the Ultra 80. However, I have can easily measure the
temperature. At the moment I have only one, but can add another 4, and
read them via the computer.

> You'd need to write a little daemon to do the parsing and job/CPU
> control as cron's resolution seems far too coarse (you care about
> 150ps, after all). pbind can be used to bind a process to a CPU,
> allowing you to control which CPU is loaded.

I care about less than 150ps actually, but that is of the time measured
by the instrument.

The errors induced in the test equipment will not occur with a change of
air inlet temperature that occur for 100's of ms or probably seconds.
Putting the CPUs on/offline will not cause rapid changes in room
temperature, so timing is not an issue.

Sorting out an algorithm might be.

> Whether this would have any effect on the room temperature remains to
> be seen, but it sounds like a fun way to waste some time if you've
> got time to waste :-)