Re: [lm-sensors] [PATCH] v1 of IBM power meter driver

["Darrick J. Wong" <djwong@us.ibm.com>]

On Tue, Aug 28, 2007 at 01:19:42PM +0200, Jean Delvare wrote:
> Hi Darrick, hi Henrique,
> 
> Good thing that manufacturers start including wattmeters in their
> hardware. Hopefully this will help users better control their power
> consumption in the long run.

Yes, we're working on that too. :)

> > What unit should we use? Watts are way, way too big as there is no
> > floating/fixed point in sysfs.  10^-6 W is probably what is called for,
> > since we already need 10^-3 V and 10^-3 A.  Small portable devices can
> > easily draw less than 10^-3 W nowadays.
> 
> Good point. The driver currently exports non-integer values, which is
> not acceptable, so indeed it needs to be changed. We want at least a

I had a feeling you might say that. :)  I'll post a follow-up patch that
presents the power meters in mW instead of W.

> resolution of 10^-3 W. Not sure about 10^-6 W. I am surprised that
> portable devices can really draw less than 1 mW, and be it the case, I

Are there meters that can measure that small a quantity of power?  If
so, then I'll vote for uW; otherwise, mW is fine enough for me.

> doubt that manufacturers will embed a wattmetter: it would probably
> draw more current than the rest of the device ;) so it may not be
> relevant for our decision.
> 
> So I think I'd go with 10^-3 W, but I welcome diverging opinions. Out
> of curiosity, what is the physical resolution of IBM's device?

I don't know for sure, but my observation is that they're no more
accurate than a Watt.  Some of those meters appear to be averages, which
would explain why they occasionally have deciWatt components.

> I see that the driver relies on IPMI. Can't it be merged with the
> out-of-tree impisensors driver then? This would give that driver some
> momentum so that it can finally be merged, and I would like to avoid
> having two drivers if one is enough. Note though that I don't know
> anything about IPMI so I might as well be totally wrong ;)

The IBM PEx meters are accessible via custom IPMI commands, not the
standard IPMI sensor interface.  I considered modifying the ipmisensors
driver, but reached the opinion that it would clutter that driver
unnecessarily, particularly since there are a lot of systems with IPMI
BMCs, but most of them will not have this interface.

That said, what is standing in the way of ipmisensors being merged?  I
think I can shake out some free time to apply TLC.

--D

Attachment: signature.asc
Description: Digital signature