National Center for Computational Sciences

Monday,
July 30, 2007 - Day 1 Morning

7:30 a.m.

Working breakfast to ensure participants have access to computer accounts to perform hands-on workshop sessions and to provide one-on-one support to discuss participant’s individual computational needs.

8:30 a.m.

Welcome, Greetings,
etc. - Doug Kothe, NCCS Director of Science (15 min) - (Video ) [WMV, 235 MB]

8:45 a.m.

Architecture of 250TF and 1000TF - (Video) [WMV, 61 MB]

• Node architecture and interconnect [PDF, 1.5 MB]- Nathan Wichman, Cray, Inc. (30 min)
• Multi-core Sockets[PDF, .7 MB] - Brian Waldecker, AMD (60 min)

10:15 a.m.

Break (15 min)

10:30 a.m.

Architecture of 250TF and 1000TF (continued)

• Inter-connect changes [PDF, 1.8 MB]- Nathan Wichmann, Cray Inc. (60
  min) - (Video) [WMV, 205 MB]

11:30 a.m.

Projecting from Today’s Application
to Larger Processor Runs [PDF, .15 MB] - John Levesque, Cray Inc.
(50 min)

12:20 p.m.

Working Lunch Catered (60 min)

• Scaling Case Study: AORSA [PDF, 2.4 MB] - Richard Barrett, NCCS

Monday,
July 30, 2007 - Day 1 Afternoon

1:20 p.m.

Case Study: Scaling Application
to 1000’s of Cores

• Scaling a Large Supernova Calculation [PDF, 2.8 MB] - Bronson Messer, NCCS (20 min) - (Video) [WMV, 148 MB]
• Combustion Simulation [PDF, 1.8 MB] - David Lignell, Sandia National Laboratories (20 min)

2:00 p.m.

Hands-On Workshop
The hands-on workshop sessions will be devoted to having NCCS/Cray
staff working closely with the attendees:

• Characterizing application bottlenecks using Cray performance
  tools
• Obtaining the best performance out of existing implementations
• Identifying alternative methods, algorithms, and infrastructure
  for better scaling

Tuesday,
July 31, 2007 - Day 2 Morning

7:30 a.m.

Working breakfast to ensure participants have access to computer accounts to perform hands-on workshop sessions and to provide one-on-one support to discuss participant’s individual computational needs.

8:30 a.m.

Projecting from Today’s
Application to Larger Processor Runs (cont’d)

• Scaling node performance and communication bottlenecks[PDF, .5 MB] - Jim Schwarzmeier, Cray Inc. (90 min) - (Video) [WMV, X MB]

10:00 a.m.

Break (15 min)

10:15 a.m.

Software for
250TF and 1000TF

• OpenMP and Pthreads on the node [PDF, .7 MB] - Ricky Kendall, NCCS
  (90 min) - (Video)[WMV, X MB]

12:15 a.m.

Working Lunch Break
(60 min)

• Cray Performance Tools [PDF, .2 MB] - Luiz DeRose, Cray

Tuesday,
July 31, 2007 - Day 2 Afternoon

1:15 p.m.

Hands-On Workshop
The hands-on workshop sessions will be devoted to having NCCS/Cray
staff working closely with the attendees:

• Characterizing application bottlenecks using Cray performance
  tools
• Obtaining the best performance out of existing implementations
• Identifying alternative methods, algorithms, and infrastructure
  for better scaling

Wednesday,
August 1, 2007 - Day 3 Morning

7:30 a.m.

Working breakfast to ensure participants have access to computer accounts to perform hands-on workshop sessions and to provide one-on-one support to discuss participant’s individual computational needs.

8:30 a.m.

NCCS Lustre Efforts [PDF, 1 MB] - Sarp Oral, NCCS (50 min) - (Video)[WMV, X MB]

9:20 a.m.

Designing
a scalable application

• [PDF, .4 MB] Applications that do not scale - Trey White, NCCS (50
  min)
• [PDF, .6 MB] Using scalable solvers - Adrian Tate, Cray Inc. (50 min)

11:00 a.m.

Break (15
min)

11:15 a.m.

Designing a scalable application
(continued)

• [PDF, .7 MB] Using Lustre Effectively - Stephen Hodson, NCCS (50 min) - (Video) [WMV, X MB]

12:00 p.m.

Working Lunch Break (60 min)

• [PDF, .6 MB] A Comparison of Application - Performance Using Open MPI and Cray MPI - Joshua Hursey, NCCS

Wednesday,
August 1, 2007 - Day 3 Afternoon

1:00 p.m.

Workshop Wrap-Up: Finalizing
Statistics and Summarizing Improvments

2:30 p.m.

Hands-On Workshop
Session
The hands-on workshop sessions will be devoted to
having NCCS/Cray staff working closely with the attendees:

• Characterizing application bottlenecks using Cray performance
  tools
• Obtaining the best performance out of existing implementations
• Identifying alternative methods, algorithms, and infrastructure
  for better scaling