Please review power management abstract for LF End User Summit

Please review power management abstract for LF End User Summit

[Brown, Len]

I've been asked to present on server power management
in a couple of weeks at the Linux Foundation End User Summit in New York.

Please review the abstract below -- I need to submit it today.

I've also penciled in an outline for the talk below,
let me know if there is something that server customers
may want to hear about that I've forgotton to include.

thanks!
-Len



Title: Topics in Linux Server Power Management
-----
Abstract:
--------
Today's high volume servers deploy several power management
features originally developed for laptop computers.
While Linux runs on laptops and thus already has support for
these features, servers have performance constraints
not generally found on laptops.

This talk starts with an overview of power management features
supported in Linux on high volumem servers, and then describes
how to maximize power savings while still meeting
server performance requirements.

No previous knowledge of Linux power management features
is necessary to benefit from this presentation.

Outline:
-------
CPU Idle Power states (C-staes)
	what they are
	how to observe
	knobs to debug/control

	C1 vs C1E
	core vs package C-states

Processor Performance States (P-state)
	what they are
	how to observe
	knobs to debug/control

	ondemand vs. userspace

Turbo Mode (Intel Dynamic Acceleration Technology)
	what it is
	how to observe
	knobs to debug/control

PM-QOS
	what it is, how it works, how to use it

Power Aware Scheduling
	what it is, how to use it

RE: Please review power management abstract for LF End User Summit

[Van De Ven, Arjan]

if you want to be cool you do a demo of the power tools we have....
makes it more interactive as well


>-----Original Message-----
>From: Brown, Len
>Sent: Wednesday, October 28, 2009 9:35
>To: Van De Ven, Arjan; Mallick, Asit K; Kleen, Andi; 'linux-power-
>mgmt@linux.intel.com'; linux-acpi@vger.kernel.org
>Subject: Please review power management abstract for LF End User Summit
>
>I've been asked to present on server power management
>in a couple of weeks at the Linux Foundation End User Summit in New
>York.
>
>Please review the abstract below -- I need to submit it today.
>
>I've also penciled in an outline for the talk below,
>let me know if there is something that server customers
>may want to hear about that I've forgotton to include.
>
>thanks!
>-Len
>
>
>
>Title: Topics in Linux Server Power Management
>-----
>Abstract:
>--------
>Today's high volume servers deploy several power management
>features originally developed for laptop computers.
>While Linux runs on laptops and thus already has support for
>these features, servers have performance constraints
>not generally found on laptops.
>
>This talk starts with an overview of power management features
>supported in Linux on high volumem servers, and then describes
>how to maximize power savings while still meeting
>server performance requirements.
>
>No previous knowledge of Linux power management features
>is necessary to benefit from this presentation.
>
>Outline:
>-------
>CPU Idle Power states (C-staes)
>	what they are
>	how to observe
>	knobs to debug/control
>
>	C1 vs C1E
>	core vs package C-states
>
>Processor Performance States (P-state)
>	what they are
>	how to observe
>	knobs to debug/control
>
>	ondemand vs. userspace
>
>Turbo Mode (Intel Dynamic Acceleration Technology)
>	what it is
>	how to observe
>	knobs to debug/control
>
>PM-QOS
>	what it is, how it works, how to use it
>
>Power Aware Scheduling
>	what it is, how to use it

Re: Please review power management abstract for LF End User Summit

[Matthew Garrett]

aspm and alpm are probably also relevant. PCI runtime pm is an 
interesting near-term development.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

RE: Please review power management abstract for LF End User Summit

[Kleen, Andi]

I would start with a power break down on a typical server.

Also at least mention things like power saving by not enabling ethernet
links.

-Andi

>-----Original Message-----
>From: Brown, Len 
>Sent: Wednesday, October 28, 2009 5:35 PM
>To: Van De Ven, Arjan; Mallick, Asit K; Kleen, Andi; 
>'linux-power-mgmt@linux.intel.com'; linux-acpi@vger.kernel.org
>Subject: Please review power management abstract for LF End User Summit
>
>I've been asked to present on server power management
>in a couple of weeks at the Linux Foundation End User Summit 
>in New York.
>
>Please review the abstract below -- I need to submit it today.
>
>I've also penciled in an outline for the talk below,
>let me know if there is something that server customers
>may want to hear about that I've forgotton to include.
>
>thanks!
>-Len
>
>
>
>Title: Topics in Linux Server Power Management
>-----
>Abstract:
>--------
>Today's high volume servers deploy several power management
>features originally developed for laptop computers.
>While Linux runs on laptops and thus already has support for
>these features, servers have performance constraints
>not generally found on laptops.
>
>This talk starts with an overview of power management features
>supported in Linux on high volumem servers, and then describes
>how to maximize power savings while still meeting
>server performance requirements.
>
>No previous knowledge of Linux power management features
>is necessary to benefit from this presentation.
>
>Outline:
>-------
>CPU Idle Power states (C-staes)
>	what they are
>	how to observe
>	knobs to debug/control
>
>	C1 vs C1E
>	core vs package C-states
>
>Processor Performance States (P-state)
>	what they are
>	how to observe
>	knobs to debug/control
>
>	ondemand vs. userspace
>
>Turbo Mode (Intel Dynamic Acceleration Technology)
>	what it is
>	how to observe
>	knobs to debug/control
>
>PM-QOS
>	what it is, how it works, how to use it
>
>Power Aware Scheduling
>	what it is, how to use it
>

Re: Please review power management abstract for LF End User Summit

[Randy Dunlap]

On Wed, 28 Oct 2009 17:07:14 +0000 Kleen, Andi wrote:

> I would start with a power break down on a typical server.
> 
> Also at least mention things like power saving by not enabling ethernet
> links.

and avoid ipmi (usually polling instead of interrupt-driven)
unless it is actively being used.

---
~Randy

RE: [Linux-power-mgmt] RE: Please review power management abstract for LF End User Summit

[Gangadhar, Mukesh]

A topic on virtualization and power management would be useful...

Thanks

Mukesh

-----Original Message-----
From: linux-power-mgmt-bounces@linux.intel.com [mailto:linux-power-mgmt-bounces@linux.intel.com] On Behalf Of Kleen, Andi
Sent: Wednesday, October 28, 2009 10:37 PM
To: Brown, Len; Van De Ven, Arjan; Mallick, Asit K; 'linux-power-mgmt@linux.intel.com'; linux-acpi@vger.kernel.org
Subject: [Linux-power-mgmt] RE: Please review power management abstract for LF End User Summit

I would start with a power break down on a typical server.

Also at least mention things like power saving by not enabling ethernet
links.

-Andi

>-----Original Message-----
>From: Brown, Len 
>Sent: Wednesday, October 28, 2009 5:35 PM
>To: Van De Ven, Arjan; Mallick, Asit K; Kleen, Andi; 
>'linux-power-mgmt@linux.intel.com'; linux-acpi@vger.kernel.org
>Subject: Please review power management abstract for LF End User Summit
>
>I've been asked to present on server power management
>in a couple of weeks at the Linux Foundation End User Summit 
>in New York.
>
>Please review the abstract below -- I need to submit it today.
>
>I've also penciled in an outline for the talk below,
>let me know if there is something that server customers
>may want to hear about that I've forgotton to include.
>
>thanks!
>-Len
>
>
>
>Title: Topics in Linux Server Power Management
>-----
>Abstract:
>--------
>Today's high volume servers deploy several power management
>features originally developed for laptop computers.
>While Linux runs on laptops and thus already has support for
>these features, servers have performance constraints
>not generally found on laptops.
>
>This talk starts with an overview of power management features
>supported in Linux on high volumem servers, and then describes
>how to maximize power savings while still meeting
>server performance requirements.
>
>No previous knowledge of Linux power management features
>is necessary to benefit from this presentation.
>
>Outline:
>-------
>CPU Idle Power states (C-staes)
>	what they are
>	how to observe
>	knobs to debug/control
>
>	C1 vs C1E
>	core vs package C-states
>
>Processor Performance States (P-state)
>	what they are
>	how to observe
>	knobs to debug/control
>
>	ondemand vs. userspace
>
>Turbo Mode (Intel Dynamic Acceleration Technology)
>	what it is
>	how to observe
>	knobs to debug/control
>
>PM-QOS
>	what it is, how it works, how to use it
>
>Power Aware Scheduling
>	what it is, how to use it
>_______________________________________________
Linux-power-mgmt mailing list
Linux-power-mgmt@linux.intel.com
http://linux.intel.com/mailman/listinfo/linux-power-mgmt

Re: Please review power management abstract for LF End User Summit

[Len Brown]

Thanks for the suggestions.

final abstract is here:
http://events.linuxfoundation.org/eus09hpc4

presentation being constructed here (currently draft 0.5):
http://userweb.kernel.org/~lenb/doc/2009-EUS-Server-PM-web/index.html

cheers,
-Len Brown, Intel Open Source Technology Center

Re: Please review power management abstract for LF End User Summit

[Robert Hancock]

On 11/06/2009 11:30 PM, Len Brown wrote:
> Thanks for the suggestions.
>
> final abstract is here:
> http://events.linuxfoundation.org/eus09hpc4
>
> presentation being constructed here (currently draft 0.5):
> http://userweb.kernel.org/~lenb/doc/2009-EUS-Server-PM-web/index.html
>
> cheers,
> -Len Brown, Intel Open Source Technology Center

I noticed you mention that "power off is a valid power management 
strategy".. indeed you can use WOL to power up again, and Linux can boot 
quickly, unfortunately the boot speed on most server platforms in the 
BIOS before even starting to boot the OS is abysmal, and seems to be 
getting worse there as it improves in desktop systems. Some IBM servers 
take MINUTES just to initialize the RAID controller.. this sort of thing 
isn't exactly conducive to rapid power-up of servers when required.

Also, the power-off power usage of some servers isn't great either - 
some of them even have to keep the power supply fans running when 
powered off, otherwise presumably the PS would heat up too much..

Presumably in both of these cases it's because of a lack of demand for 
better behavior (fast power-up on desktops is demanded by Microsoft 
"designed for Windows" specs, and low power-off power usage is required 
by Energy Star, EU standards, etc.)

Re: Please review power management abstract for LF End User Summit

[Len Brown]

On Sat, 14 Nov 2009, Robert Hancock wrote:

> On 11/06/2009 11:30 PM, Len Brown wrote:
> > Thanks for the suggestions.
> > 
> > final abstract is here:
> > http://events.linuxfoundation.org/eus09hpc4
> > 
> > presentation being constructed here (currently draft 0.5):
> > http://userweb.kernel.org/~lenb/doc/2009-EUS-Server-PM-web/index.html

> I noticed you mention that "power off is a valid power management strategy"..
> indeed you can use WOL to power up again, and Linux can boot quickly,
> unfortunately the boot speed on most server platforms in the BIOS before even
> starting to boot the OS is abysmal, and seems to be getting worse there as it
> improves in desktop systems. Some IBM servers take MINUTES just to initialize
> the RAID controller.. this sort of thing isn't exactly conducive to rapid
> power-up of servers when required.

Agreed, though not all systems have these delays -- choose wisely:-)

> Also, the power-off power usage of some servers isn't great either - some of
> them even have to keep the power supply fans running when powered off,
> otherwise presumably the PS would heat up too much..

ditto.

> Presumably in both of these cases it's because of a lack of demand for better
> behavior (fast power-up on desktops is demanded by Microsoft "designed for
> Windows" specs, and low power-off power usage is required by Energy Star, EU
> standards, etc.)

Yes, I agree think that these external forces are effective in getting
system vendors in line.  Another is formal RFQ's for large purchases
where the customer specifies exactly what they want.

cheers,
-Len