RE: [eman] Comments on draft-parello-eman-definitions-03

Bruce Nordman <bnordman@lbl.gov><= /p> · Mon, 14 Nov 2011 02:48:20 = -0800

_______________________________________________
eman mailing list
eman@ietf.org
https://www.ietf.org/mailman/listinfo/eman

5.2.3.  Multiple power state sets   The energy management standard must provide means for supporting
   multiple power state sets simultaneously at a powered entity.
On Thu, Nov 10, 2011 at 11:01:58AM -0600, Mielke, William F (Bill) wrote:

Comments provided inline below.

From: Juergen Schoenwaelder 
[mailto:j.schoenwaelder@jacobs-university.de] 
Sent: Thursday, November 10, 2011 4:25 AM
To: Mielke, William F (Bill)
Cc: eman@ietf.org
Subject: Re: [eman] iana powerstate tc proposal

What is "behave reasonably"? If we can't define the mappings, how are
implementors going to figure this out? And surely, implementations
will exhibit different ideas what "behave reasonably" means. For me,
the poor devices are really the wrong place to please the managers who
prefer to live a world that has only one set of power states.

Anyway, if the WG really is eager to go down this path, there should
in my view very clear words telling implementors "do not cheat with
your interpretation of behave reasonably" and there should be clear
words how to react if a "behave reasonably" mapping does not exist,
e.g. to fail any set requests that fall out of the "behave reasonably"
space for some definition of "behave reasonably".

I am actually in agreement with this point which is why I raised the issue.  We must provide some guidance on how the devices are expected to behave if they support multiple powerstate sets concurrently.  I fully agree that "behave reasonably" is insufficient and vague which is why I attempted to provide some characterization of what the problem was.

I believe that the definition of "behave reasonably" is likely to be technology and perhaps even device specific which is why we have been thus far unable to articulate a universal mapping.  In the absence of such a universal mapping what are we to do?  Say nothing at all?  I believe that we need to push through and provide some level of guidance, even if that guidance is known to be imperfect.  But that is my opinion.

I would be OK with also providing the general guidance that in the event that a given device cannot devise a "reasonable" mapping and set of behaviors to insure that different managers acting independently each receive a set of responses which are consistent with their respective interfaces that we simply recommend that they not attempt to support multiple powerstate sets.

I think the burden of coordination should be on the manager side. We
have many MIB objects in SNMP land where uncoordinated managers will
screw things up. So this is nothing new or specific to EMAN. And if
two uncoordinated manages like to exercise power management by
modifying power states on a device without knowing of each other, I
believe you are likely screwed even if both stay in the same set of
power states or if you allow concurrent power states. Coordination
really is a separate issue.

I also agree with Juergen that the scenario I raise of having two uncoordinated managers both attempting to control powerstates is almost guaranteed to lead to chaos, regardless of whether they are using the same or different powerstate sets.  But we must work within the framework we have.  Given the device centric nature of SNMP we are naturally driven to device centric solutions, or at least half solutions.

I am still relatively new to IETF.  Are there accepted mechanisms within IETF whereby we can constrain what the managers are allowed to do?  Are there mechanisms by which we can force the managers to coordinate their activities?  If so I would be open to exploring those as well to help insure that what we define is as rigorous as possible.

Are there any examples of how other working groups have approached this type of issue?

SNMP has an access control model - you can configure it such that only
one manager has write access. This forces managers to coordinate with
each other. (But if you do this, supporting multiple concurrent power
states _on the device_ becomes a mood point.)

Except access control, SNMP does not provide a generic coordination
service. The SNMP approach has always been that agents should be
simple, following the general idea that complex functions are cheaper
to implement on managers than on agents embedded on devices.

That said, there are a few MIB modules where functions are maintained
for potentially multiple managers. The typical solution is to tag rows
such that managers get different "sandboxes" to play with. This kind
of works if the functions are mostly independent. Unfortunately, a
power state is bound to the device itself and thus sandboxing does not
work.

/js

5.2.3.  Multiple power state sets   The energy manageme=
nt standard must provide means for supporting
   multiple power state sets simultaneously at a powered entity.
_______________________________________________
eman mailing list
eman@ietf.org
https://www.ietf.org/mailman/listinfo/eman

   The energy management =
standard must provide means for =
supporting
   multiple power state sets =
simultaneously at a powered entity.
=A0=A0 The energy management standard must provide means for supportin=
g
=A0=A0 multiple power state sets simultaneously a=
t a powered entity.
   The energy management standard must provide means for supporting
   multiple power state sets simultaneously at a powered entity.
_______________________________________________
eman mailing list
eman@ietf.org
https://www.ietf.org/mailman/listinfo/eman

_______________________________________________
eman mailing list
eman@ietf.org
https://www.ietf.org/mailman/listinfo/eman

Subject:	Re: [eman] iana = powerstate tc proposal
Date: =	Mon, 14 Nov 2011 02:48:20 = -0800
From:	Bruce Nordman <bnordman@lbl.gov><= /p>
To:	Benoit Claise <bclaise@cisco.com>
CC:	Mielke, William F = (Bill) <bill.mielke@alcatel-lu= cent.com>, eman@ietf.org <eman@ietf.org>

Subject:	Re: [eman] iana powerstate tc proposal=
Dat= e:	Mon, 14 N= ov 2011 02:48:20 -0800
From:	<= p class=3D"MsoNormal">Bruce Nordman <bnordman@lbl.gov>
To:	Benoit Claise <bclaise@cisco.com>
<= p class=3D"MsoNormal" style=3D"text-align:right" align=3D"right">CC: =	Mielke, William F (Bill) <bill.mielke@alcatel-lucent.com>, eman@ietf.org <eman@ietf.org>

Subject:	Re: [eman] iana powerstate tc proposal
Date:	Mon, 14 Nov 2011 02:48:20 -0800
From:	Bruce Nordman <bnordman@lbl.gov>
To:	Benoit Claise <bclaise@cisco.com>
CC:	Mielke, William F (Bill) <bill.mielke@alcatel-lucent.com>, eman@ietf.org <eman@ietf.org>

5.2.3.=A0 Multiple power stat= e sets

5.2.3. Multiple power state sets

5.2.3. Multiple power state sets

5.2.3. Multiple power state sets

5.2.3. Multiple power state = sets