power draw depending on PHY speed

power draw depending on PHY speed

[Leisner, Martin]

While doing some current probe measurements (i.e. to measure the effects
of NAP and frequency scaling with a freescale processor) I noticed some
very
significant differents if we ran on 100Mbit of 1000Mbit ethernet.

The hardware guys like at their data sheets and agreed.

Does anyone "drop ethernet speed" to save power?
(when a system is idle, is isn't necessary to run 1G ethernet).


When a system has WOL capability, what speed does it run at when the
system
is sleeping?

marty

Re: [linux-pm] power draw depending on PHY speed

[Rafael J. Wysocki]

On Friday 28 August 2009, Leisner, Martin wrote:
> While doing some current probe measurements (i.e. to measure the effects
> of NAP and frequency scaling with a freescale processor) I noticed some
> very
> significant differents if we ran on 100Mbit of 1000Mbit ethernet.
> 
> The hardware guys like at their data sheets and agreed.
> 
> Does anyone "drop ethernet speed" to save power?
> (when a system is idle, is isn't necessary to run 1G ethernet).

I don't know, really.  This is a netdev question IMO (CC added).

> When a system has WOL capability, what speed does it run at when the
> system is sleeping?

That's a very good question.  I bet that depends on the NIC in question, at
least I'm not sure if any spec regulates it.

Thanks,
Rafael

Re: [linux-pm] power draw depending on PHY speed

[Ben Hutchings]

On Fri, 2009-08-28 at 21:03 +0200, Rafael J. Wysocki wrote:
> On Friday 28 August 2009, Leisner, Martin wrote:
> > While doing some current probe measurements (i.e. to measure the effects
> > of NAP and frequency scaling with a freescale processor) I noticed some
> > very
> > significant differents if we ran on 100Mbit of 1000Mbit ethernet.
> > 
> > The hardware guys like at their data sheets and agreed.
> > 
> > Does anyone "drop ethernet speed" to save power?
> > (when a system is idle, is isn't necessary to run 1G ethernet).
> 
> I don't know, really.  This is a netdev question IMO (CC added).

Doing this through the existing autonegotiation mechanism is a bad idea,
because it requires bringing the link down, possibly for a few seconds.
There is a standard on the way (802.3az, aka Energy-Efficient Ethernet
or EEE) which will allow for speed renegotiation without dropping the
link.

> > When a system has WOL capability, what speed does it run at when the
> > system is sleeping?
> 
> That's a very good question.  I bet that depends on the NIC in question, at
> least I'm not sure if any spec regulates it.

I'm not aware of a spec, but generally a 1G or 10G multi-speed PHY will
not advertise speeds higher than 100 Mbit/s when sleeping.

Ben.

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

Re: power draw depending on PHY speed

[Ben Hutchings]

On Fri, 2009-08-28 at 21:03 +0200, Rafael J. Wysocki wrote:
> On Friday 28 August 2009, Leisner, Martin wrote:
> > While doing some current probe measurements (i.e. to measure the effects
> > of NAP and frequency scaling with a freescale processor) I noticed some
> > very
> > significant differents if we ran on 100Mbit of 1000Mbit ethernet.
> > 
> > The hardware guys like at their data sheets and agreed.
> > 
> > Does anyone "drop ethernet speed" to save power?
> > (when a system is idle, is isn't necessary to run 1G ethernet).
> 
> I don't know, really.  This is a netdev question IMO (CC added).

Doing this through the existing autonegotiation mechanism is a bad idea,
because it requires bringing the link down, possibly for a few seconds.
There is a standard on the way (802.3az, aka Energy-Efficient Ethernet
or EEE) which will allow for speed renegotiation without dropping the
link.

> > When a system has WOL capability, what speed does it run at when the
> > system is sleeping?
> 
> That's a very good question.  I bet that depends on the NIC in question, at
> least I'm not sure if any spec regulates it.

I'm not aware of a spec, but generally a 1G or 10G multi-speed PHY will
not advertise speeds higher than 100 Mbit/s when sleeping.

Ben.

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

Re: power draw depending on PHY speed

[Rafael J. Wysocki]

On Friday 28 August 2009, Leisner, Martin wrote:
> While doing some current probe measurements (i.e. to measure the effects
> of NAP and frequency scaling with a freescale processor) I noticed some
> very
> significant differents if we ran on 100Mbit of 1000Mbit ethernet.
> 
> The hardware guys like at their data sheets and agreed.
> 
> Does anyone "drop ethernet speed" to save power?
> (when a system is idle, is isn't necessary to run 1G ethernet).

I don't know, really.  This is a netdev question IMO (CC added).

> When a system has WOL capability, what speed does it run at when the
> system is sleeping?

That's a very good question.  I bet that depends on the NIC in question, at
least I'm not sure if any spec regulates it.

Thanks,
Rafael

Re: power draw depending on PHY speed

[Pavel Machek]

On Fri 2009-08-28 08:26:13, Leisner, Martin wrote:
> While doing some current probe measurements (i.e. to measure the effects
> of NAP and frequency scaling with a freescale processor) I noticed some
> very
> significant differents if we ran on 100Mbit of 1000Mbit ethernet.

Yep, difference seems to be cca 1W. I had stupid shell script to switch it...

								Pavel


-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html