From doremire at ucar.edu  Fri Mar  2 12:44:01 2007
From: doremire at ucar.edu (Emily Doremire)
Date: Fri, 02 Mar 2007 12:44:01 -0700
Subject: [ccsm-scientists] CCSM Scientists Meeting 2pm today, 3/2/07
Message-ID: <45E87E81.4030605@ucar.edu>

Just as a reminder, there will be a CCSM scientists meeting today, 
3/2/07, at 2 pm Mountain Time.  No presentations are expected, so there 
will not be a webcast.  If you wish to join in remotely, please dial in 
to the conference call.

   * Dial in to either of these numbers:  1-800-516-9896 or 1-816-650-0725.
   * When prompted, enter participant code 373971.  You will be placed 
on hold until the chairperson begins the call.

-- 
Emily Doremire
NCAR CGD CCSM
P.O. Box 3000
Boulder, CO 80307-3000
303-497-8660 phone
303-497-1314 fax



From gent at cgd.ucar.edu  Wed Mar  7 09:51:23 2007
From: gent at cgd.ucar.edu (PETER GENT)
Date: Wed, 7 Mar 2007 09:51:23 -0700 (MST)
Subject: [ccsm-scientists] CSL/CCSM Allocations
Message-ID: <Pine.LNX.4.64.0703070948160.19432@gentoo2.cgd.ucar.edu>

---------- Forwarded message ----------
Date: Tue, 06 Mar 2007 06:00:58 -0700
From: Tom Bettge <bettge at ucar.edu>
To: PETER GENT <gent at cgd.ucar.edu>
Cc:  <cal at ucar.edu>, Jon Wolfe <jwolfe at cgd.ucar.edu>,
    Mariana Vertenstein <mvertens at cgd.ucar.edu> 
Subject: Re: CSL/CCSM Allocations

Peter:

I understand your concerns.  We do face several challenges during the
transition period - balancing usage on the new machine during the
so-called special project time, assuring access to NCAR, Univ, and CSL
users, the recalibration of the charge factors as CCSM has pointed out.

I'll get to the bottom line........as a result, we are increasing the
hold thresholds for both CCSM Dev and Prod by 10%, effectively
immediately.  We will reexamine the available GAUs for distribution on
April 1 (no fooling), and see what we can do then.  If CCSM projects
find that they cannot finish important work as the CCSM workshop
approaches, we will do all we can to find resources to help - I would
ask that they direct their requests through you to Ginger in order that
you have the opportunity to prioritize the requests.

Back to other concerns, just FYI......

The change in the charge factors was based upon benchmark results which
CISL staff (not IBM) ran during the January testing period.  Thus far,
based upon the metrics we observe, including average job efficiency in
terms of % of peak, combined with the known clock increase from bluesky
to bluevista and blueice, we have seen that the 0.87 charge factor is
generally correct compared to bluesky.  What does this mean?  First, it
would mean that the workloads are identical, and I'm certain that is not
true.  Second, not all applications will transition at the same
performance level as others, and clearly CCSM is one which suffers.  The
CISL staff is currently assisting CSEG to determine what changes can be
incorporated to improve the performance - we are certainly committed to
helping out, and the onsite IBM applications specialist Jim Edwards is
also working on this.  I assume that changes in compiler optimizations,
clock rates, switch bandwidth and latency, which would affect components
differently, would also call for a task re-balance within each CCSM
configuration, and perhaps that has already been done.

We increased the GAUs to NCAR projects, and not to CSL, during the
spring for a couple of reasons.....first, the allocations given to the
special projects were directed mainly toward the non-NCAR community so
we felt that the NCAR community should benefit; in addition, NSF
decreased funding to the CSL in FY07 by $680K.  The second point will
also be seen/felt during the reallocation of all resources on Jun 1
(that date has changed from May 1, as approved by the NCAR Director's
Office), and later in the year during the CSL reallocation.

Please let me know if you have any questions....you are always welcome
to stop by as well!

..Tom


PETER GENT wrote:
> Tom and Ginger, 
>
> I have been told by Jon, Mariana and others that two factors have 
> contributed to a substantial decrease in the value of CCSM's CSL time. 
> The first is that CCSM is costing ~20% more GAUs to run on the IBMs 
> remaining, than it did on Bluesky.  The second is the change in cost 
> factor from 0.63 to 0.87 on Blueice.  This has been compensated by a 
> 40% increase in allocation for many users, but not CSL allocations. 
> Together, these mean ~30% decrease in CSL time over March and April, 
> before the large CSL allocation increase starts on 1 May.  It would 
> be very difficult to overcome this decrease either by a change in 
> priorities, or by efficiencies within the CCSM code, so this decrease 
> will negatively affect CCSM progress.  Is there any possibility of 
> temporarily increasing the CSL/CCSM allocations over March and April? 
>
> Thanks,  Peter.

From wily at ucar.edu  Wed Mar  7 10:20:48 2007
From: wily at ucar.edu (W. G. Large)
Date: Wed, 7 Mar 2007 10:20:48 -0700 (MST)
Subject: [ccsm-scientists] CSL/CCSM Allocations
In-Reply-To: <Pine.LNX.4.64.0703070948160.19432@gentoo2.cgd.ucar.edu>
References: <Pine.LNX.4.64.0703070948160.19432@gentoo2.cgd.ucar.edu>
Message-ID: <Pine.LNX.4.64.0703071018400.30939@jackson.cgd.ucar.edu>

On Wed, 7 Mar 2007, PETER GENT wrote:

PETER :
        It is unclear from below if the June 1 
reallocation will  still be until 31 Oct. (5 months)
or for 6 months.

I guess we all have to sacrifice for "breakthrough" science !!!

                            BILL

> ---------- Forwarded message ----------
> Date: Tue, 06 Mar 2007 06:00:58 -0700
> From: Tom Bettge <bettge at ucar.edu>
> To: PETER GENT <gent at cgd.ucar.edu>
> Cc:  <cal at ucar.edu>, Jon Wolfe <jwolfe at cgd.ucar.edu>,
>     Mariana Vertenstein <mvertens at cgd.ucar.edu> 
> Subject: Re: CSL/CCSM Allocations
> 
> Peter:
> 
> I understand your concerns.  We do face several challenges during the
> transition period - balancing usage on the new machine during the
> so-called special project time, assuring access to NCAR, Univ, and CSL
> users, the recalibration of the charge factors as CCSM has pointed out.
> 
> I'll get to the bottom line........as a result, we are increasing the
> hold thresholds for both CCSM Dev and Prod by 10%, effectively
> immediately.  We will reexamine the available GAUs for distribution on
> April 1 (no fooling), and see what we can do then.  If CCSM projects
> find that they cannot finish important work as the CCSM workshop
> approaches, we will do all we can to find resources to help - I would
> ask that they direct their requests through you to Ginger in order that
> you have the opportunity to prioritize the requests.
> 
> Back to other concerns, just FYI......
> 
> The change in the charge factors was based upon benchmark results which
> CISL staff (not IBM) ran during the January testing period.  Thus far,
> based upon the metrics we observe, including average job efficiency in
> terms of % of peak, combined with the known clock increase from bluesky
> to bluevista and blueice, we have seen that the 0.87 charge factor is
> generally correct compared to bluesky.  What does this mean?  First, it
> would mean that the workloads are identical, and I'm certain that is not
> true.  Second, not all applications will transition at the same
> performance level as others, and clearly CCSM is one which suffers.  The
> CISL staff is currently assisting CSEG to determine what changes can be
> incorporated to improve the performance - we are certainly committed to
> helping out, and the onsite IBM applications specialist Jim Edwards is
> also working on this.  I assume that changes in compiler optimizations,
> clock rates, switch bandwidth and latency, which would affect components
> differently, would also call for a task re-balance within each CCSM
> configuration, and perhaps that has already been done.
> 
> We increased the GAUs to NCAR projects, and not to CSL, during the
> spring for a couple of reasons.....first, the allocations given to the
> special projects were directed mainly toward the non-NCAR community so
> we felt that the NCAR community should benefit; in addition, NSF
> decreased funding to the CSL in FY07 by $680K.  The second point will
> also be seen/felt during the reallocation of all resources on Jun 1
> (that date has changed from May 1, as approved by the NCAR Director's
> Office), and later in the year during the CSL reallocation.
> 
> Please let me know if you have any questions....you are always welcome
> to stop by as well!
> 
> ..Tom
> 
> 
> PETER GENT wrote:
> > Tom and Ginger, 
> >
> > I have been told by Jon, Mariana and others that two factors have 
> > contributed to a substantial decrease in the value of CCSM's CSL time. 
> > The first is that CCSM is costing ~20% more GAUs to run on the IBMs 
> > remaining, than it did on Bluesky.  The second is the change in cost 
> > factor from 0.63 to 0.87 on Blueice.  This has been compensated by a 
> > 40% increase in allocation for many users, but not CSL allocations. 
> > Together, these mean ~30% decrease in CSL time over March and April, 
> > before the large CSL allocation increase starts on 1 May.  It would 
> > be very difficult to overcome this decrease either by a change in 
> > priorities, or by efficiencies within the CCSM code, so this decrease 
> > will negatively affect CCSM progress.  Is there any possibility of 
> > temporarily increasing the CSL/CCSM allocations over March and April? 
> >
> > Thanks,  Peter.
> _______________________________________________
> ccsm-scientists mailing list
> ccsm-scientists at cgd.ucar.edu
> http://www.cgd.ucar.edu/mailman/listinfo/ccsm-scientists
> 

From gent at cgd.ucar.edu  Fri Mar 16 11:52:30 2007
From: gent at cgd.ucar.edu (PETER GENT)
Date: Fri, 16 Mar 2007 11:52:30 -0600 (MDT)
Subject: [ccsm-scientists] Meeting at 2pm
Message-ID: <Pine.LNX.4.64.0703161148500.17767@gentoo2.cgd.ucar.edu>

All, 

I've decided to cancel the CCSM Scientist meeting scheduled for this 
afternoon at 2pm.  The Software Engineering WG meeting is going on, 
and several NCAR Cochairs are away as well.  The next meeting is 
scheduled for Thursday March 29th at 2pm, I believe, 

Cheers,  Peter. 

From doremire at ucar.edu  Thu Mar 29 13:03:02 2007
From: doremire at ucar.edu (Emily Doremire)
Date: Thu, 29 Mar 2007 13:03:02 -0600
Subject: [ccsm-scientists] CCSM Scientists' Meeting today, 3/29/07,
	at 2pm in ML Daomon Room
Message-ID: <460C0D66.9060608@ucar.edu>

Good afternoon everyone,

There will be a CCSM scientists' meeting today at 2pm in the ML Damon
Room.  Updates on fully coupled runs with 3.5 CAM, CLM, and CICE by
David Bailey and aerosol data sets for 1870-2000 and CAM runs by Phil
Rasch are on the agenda.

For those of you who would like to join the meeting remotely, there will
be a webcast of the presentations and conference call available.

For the webcast:

   * Open a java enabled web-browser. Mozilla based browsers are best
(i.e. Firefox, Netscape, Mozilla).
   * Go to the following URL: http://whiteboard.cgd.ucar.edu
   * Type in the password "irac96%#" without the quotes.

For the conference call:

   * Dial in to either of these numbers:  1-800-516-9896 or 1-816-650-0725.
   * When prompted, enter participant code 373971.  You will be placed
on hold until the chairperson begins the call.

-- 
Emily Doremire
UCAR NCAR CGD CCSM
1850 Table Mesa Drive
Boulder, CO 80305
303-497-8660 phone
303-497-1314 fax