From lydia at ucar.edu Tue Jan 17 14:08:29 2006 From: lydia at ucar.edu (Lydia Shiver) Date: Tue Jan 17 14:08:32 2006 Subject: [CCSM-SEWG] CCSM SEWG Agenda Message-ID: <43CD5CCD.1080908@ucar.edu> Hello everyone, The agenda for the SEWG meeting scheduled on Wed., Jan. 25, 2006 can be viewed at http://www.ccsm.ucar.edu/working_groups/Software/agendas/060125.pdf. If your group is attending the SEWG via Access Grid, please pass this agenda on to your AG operator and ask them to contact Darin Oman, NCAR's AG operator at darin@ucar.edu. If you have not let me know that you will attend via being at NCAR, Access Grid, or Webcast, please do so. Attendees that I know about are: NCAR Access Grid ---------------- Mariana Vertenstein, co-chair Pat Worley, co-chair Cecelia DeLuca, co-chair Forrest Hoffman, ORNL Nancy Norton Erik Kluzek Mat Rothstein Peter Gent John Truesdale Lawrence Buja Brian Kauffman Brian Eaton ORNL AG ------- John Drake LBL AG ------ Helen He Yu-Heng LLNL AG ------- Art Mirin LANL AG ------- Phil Jones Argonne AG ---------- Jay Larson Ray Loy Rob Jacob Thanks, Lydia From yhe at lbl.gov Mon Jan 23 12:07:38 2006 From: yhe at lbl.gov (yunhe) Date: Mon Jan 23 14:25:55 2006 Subject: [CCSM-SEWG] Re: schedule for SEWG meeting In-Reply-To: <43C6F570.7040304@ucar.edu> References: <43C6F570.7040304@ucar.edu> Message-ID: <43D5297A.5030404@lbl.gov> Hi, Mariana, Pat, and SEWGers, I asked Mariana at the BGC telecon this morning. Do you have any insight about this CAM performance run result? Always heard that CAM has large memory bandwidth requirement, but recent tests on various architectures by LBL benchmarking team (separate from our SciDAC team) with single core and dual core has only resulted 2% performance penalty with dual core, which supposedly with memory contention due to dual core, we should see bigger penalty. So, was CAM actually latency dominated? (also heard CAM large memory footprint requirement). Any other explanations? Thank you very much! Helen From lydia at ucar.edu Mon Jan 23 16:49:59 2006 From: lydia at ucar.edu (Lydia Shiver) Date: Mon Jan 23 16:50:03 2006 Subject: [CCSM-SEWG] CCSM Software Engineering Working Group Meeting Message-ID: <43D56BA7.8000609@ucar.edu> Hello everyone, The CCSM Software Engineering Working Group (SEWG) meeting is scheduled for this Wednesday, January 25, 2006 at the Vislab at NCAR. The agenda is posted on the Web and can be viewed at http://www.ccsm.ucar.edu/working_groups/Software/agendas/060125.pdf. If you are participating via the Webcast, please go to www.ccsm.ucar.edu/vnc and the password is 1rac96%#. The conference call number is 1-800-516-9896, and the participant code is 373971. If you plan on attending via the Access Grid, and have not already done so, please ask your access grid operator to contact Darin Oman, NCAR's AG operator at darin@ucar.edu. If is very important that Carol Wimert (cwimert@ucar.edu) receive all of your Powerpoint presentations for the Access Grid presentation by noon on Tuesday, Jan. 24. Thank you, Lydia and Carol From worleyph at ornl.gov Mon Jan 23 20:29:18 2006 From: worleyph at ornl.gov (Patrick H. Worley) Date: Tue Jan 24 08:37:41 2006 Subject: [CCSM-SEWG] Re: schedule for SEWG meeting In-Reply-To: <43D5297A.5030404@lbl.gov> References: <43C6F570.7040304@ucar.edu> <43D5297A.5030404@lbl.gov> Message-ID: <43D59F0E.4060701@ornl.gov> yunhe wrote: > Hi, Mariana, Pat, and SEWGers, > > I asked Mariana at the BGC telecon this morning. Do you have any > insight about this CAM performance run result? > > Always heard that CAM has large memory bandwidth requirement, but > recent tests on various architectures by LBL benchmarking team > (separate from our SciDAC team) with single core and dual core has > only resulted 2% performance penalty with dual core, which supposedly > with memory contention due to dual core, we should see bigger penalty. > So, was CAM actually latency dominated? (also heard CAM large memory > footprint requirement). Any other explanations? > > Thank you very much! > Helen > > > > There is not enough information here - what dycore, what problem size? For FV dynamics, memory bandwidth requirements are smaller than for spectral dycore. Also, setting PCOLS smaller cuts down on the memory contention. So, things are "better" than in the CCM days. However, once all of the tracers for atmospheric chemistry are turned, memory contention may become more of a problem again. Pat From gcarr at ucar.edu Tue Jan 24 09:54:11 2006 From: gcarr at ucar.edu (George R Carr Jr) Date: Tue Jan 24 10:03:12 2006 Subject: [CCSM-SEWG] Re: schedule for SEWG meeting In-Reply-To: <43D5297A.5030404@lbl.gov> References: <43C6F570.7040304@ucar.edu> <43D5297A.5030404@lbl.gov> Message-ID: At 11:07 AM -0800 1/23/06, yunhe wrote: >Hi, Mariana, Pat, and SEWGers, > >I asked Mariana at the BGC telecon this morning. Do you have any >insight about this CAM performance run result? > >Always heard that CAM has large memory bandwidth requirement, but >recent tests on various architectures by LBL benchmarking team >(separate from our SciDAC team) with single core and dual core has >only resulted 2% performance penalty with dual core, which >supposedly with memory contention due to dual core, we should see >bigger penalty. So, was CAM actually latency dominated? (also heard >CAM large memory footprint requirement). Any other explanations? > >Thank you very much! >Helen > > > >_______________________________________________ >CCSM-SEWG mailing list >CCSM-SEWG@cgd.ucar.edu >http://www.cgd.ucar.edu/mailman/listinfo/ccsm-sewg The results are very dependent on the machines you are testing. I'd need to see more of the machine and application configurations to speculate much about any inferences. On NCAR's IBM and Xeon machines, memory bandwidth limitations are definitely indicated. The runs with Tau on the NCAR IBM clearly indicate this. -- =========================== George R Carr Jr http://www.ccsm.ucar.edu/ From YHe at lbl.gov Tue Jan 24 20:47:47 2006 From: YHe at lbl.gov (Helen He) Date: Wed Jan 25 07:38:55 2006 Subject: [CCSM-SEWG] Re: schedule for SEWG meeting Message-ID: <1b4fd61aec47.1aec471b4fd6@lbl.gov> Pat and George, Thank you very much for your replies. Here is some info I obtained from the LBL benchmarking team: ======================================================================= * Spectral core * T85 problem size * Power5+ 1.9GHz with dual core processors (pre-release systems). The result is also reproducible on dual-core AMD systems for a much smaller scale problem. ====================================================================== Would this info be helpful for more explanation from you? Thanks! Helen > The results are very dependent on the machines you are testing. > I'd need to > see more of the machine and application configurations to > speculate much about > any inferences. On NCAR's IBM and Xeon machines, memory bandwidth > limitationsare definitely indicated. The runs with Tau on the NCAR > IBM clearly indicate > this. > -- > George R Carr Jr > There is not enough information here - what dycore, what problem size? > For FV dynamics, memory bandwidth requirements are smaller than > for spectral dycore. Also, setting PCOLS smaller cuts down on the > memory contention. So, things are "better" than in the CCM days. > However, once all of the tracers for atmospheric chemistry > are turned, memory contention may become more of > a problem again. > > Pat From erik at ucar.edu Wed Jan 25 11:22:21 2006 From: erik at ucar.edu (Erik Kluzek) Date: Wed Jan 25 11:31:59 2006 Subject: [CCSM-SEWG] SWIKI pages moved... In-Reply-To: <1b4fd61aec47.1aec471b4fd6@lbl.gov> References: <1b4fd61aec47.1aec471b4fd6@lbl.gov> Message-ID: Note to everyone... I moved the SWIKI pages to the new CCSM SWIKI. The base of the CCSM SWIKI is http://swiki.ucar.edu/ccsm/ Helen's SINGLE Exec page is under... http://swiki.ucar.edu/ccsm/11 Erik Kluzek, (CGD at NCAR) National Center for Atmospheric Research Boulder CO, (off) (303)497-1326 (fax) (303)497-1314 ------------------ Home page ------------------------ http://www.cgd.ucar.edu/~erik !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!