Difference between revisions of "OWG Meeting 1-Apr-2009"
m |
m |
||
Line 25: | Line 25: | ||
==Announcements== | ==Announcements== | ||
+ | DAQ group just hired Bryan Moffit to help Dave Abbott with embedded computers and related things. He will start mid-April. | ||
Line 39: | Line 40: | ||
=Minutes= | =Minutes= | ||
− | Attendees: | + | Attendees: Elliott W, Beni Z, David L, Simon T, Chris C, Dave A, Alex S, Elton S, Eugene C, Vardan G, Carl T, Dave D, Fernando B, Ed J, Jim S. |
+ | |||
+ | |||
+ | '''Online Project List''' | ||
+ | |||
+ | Elliott briefly showed the project list, and there was a small amount of discussion. The list currently is just a core-dump of projects. Elliott needs to organize and prioritize items on the list, and add "need by" dates. | ||
+ | |||
+ | |||
+ | '''CHEP Report''' | ||
+ | |||
+ | Carl, David L and Vardan gave an informal report on the CHEP conference in Prague. | ||
+ | |||
+ | * Dominated by LHC talks. | ||
+ | * Python widely used for scripting, even parallel Python. | ||
+ | * C++ problems noted, program bloat, performance problems due to bad code. | ||
+ | * Many performance monitoring tools being used. | ||
+ | * Multi-core processors becoming more important, up to 96 cores. Most codes NOT threaded and optimized for multi-core. What are the benefits of a multi-threaded program vs multiple jobs on the same multi-core processor? | ||
+ | * Cloud computing was a hot topic. 20 cents per hour per cpu on commercial farm. Virtual machines used, external I/O slow (5 MB/s max). | ||
+ | * Some LHC experiments use web for run control. | ||
+ | * ROOT widely used for analysis. Getting new C++ interpreter (LLBM) and switching to OpenGL. | ||
+ | * Data center problems: power, cooling. Old tankers even being used to house farms! | ||
+ | * Some talks on machine learning and advanced algorithms. | ||
+ | * CMS using XDAQ - 55M channels. XDAQ took 8 people 8 years = 64 M-Y. | ||
+ | * C++ reflection package from CERN. | ||
+ | * PROOF widely used. | ||
+ | * Starting runs at LHC slow. One experiment has 2000 nodes, 10000 apps. | ||
+ | * ATLAS uses Java event display. | ||
+ | * GPU's (processors) and CUDA (language) being explored. | ||
+ | * Virtualization may cause as little as 6% loss of cpu. | ||
+ | * Data preservation - plan for this from the beginning. | ||
+ | * Scientific Linux very popular everywhere but JLab. |
Revision as of 16:01, 1 April 2009
Contents
Agenda
- Announcements -
- Review of minutes from 18-Mar-2009 meeting
- Online Project List - Elliott
- Report from CHEP - David, Carl, Vardan
- Solenoid coil test - Elliott
- Electronics status reports - Fernando, Chris, others
Time/Location
1:30 PM Wed 1-Apr-2009 CC F326
To connect by telephone:
1.) dial:
- 800-377-8846 : US
- 888-276-7715 : Canada
- 302-709-8424 : International
2.) enter participant code: 39527048# (remember the "#")
Announcements
DAQ group just hired Bryan Moffit to help Dave Abbott with embedded computers and related things. He will start mid-April.
Next Meeting
15-Apr-2009 1:30pm CC F326
New Action Items from this Meeting
Minutes
Attendees: Elliott W, Beni Z, David L, Simon T, Chris C, Dave A, Alex S, Elton S, Eugene C, Vardan G, Carl T, Dave D, Fernando B, Ed J, Jim S.
Online Project List
Elliott briefly showed the project list, and there was a small amount of discussion. The list currently is just a core-dump of projects. Elliott needs to organize and prioritize items on the list, and add "need by" dates.
CHEP Report
Carl, David L and Vardan gave an informal report on the CHEP conference in Prague.
- Dominated by LHC talks.
- Python widely used for scripting, even parallel Python.
- C++ problems noted, program bloat, performance problems due to bad code.
- Many performance monitoring tools being used.
- Multi-core processors becoming more important, up to 96 cores. Most codes NOT threaded and optimized for multi-core. What are the benefits of a multi-threaded program vs multiple jobs on the same multi-core processor?
- Cloud computing was a hot topic. 20 cents per hour per cpu on commercial farm. Virtual machines used, external I/O slow (5 MB/s max).
- Some LHC experiments use web for run control.
- ROOT widely used for analysis. Getting new C++ interpreter (LLBM) and switching to OpenGL.
- Data center problems: power, cooling. Old tankers even being used to house farms!
- Some talks on machine learning and advanced algorithms.
- CMS using XDAQ - 55M channels. XDAQ took 8 people 8 years = 64 M-Y.
- C++ reflection package from CERN.
- PROOF widely used.
- Starting runs at LHC slow. One experiment has 2000 nodes, 10000 apps.
- ATLAS uses Java event display.
- GPU's (processors) and CUDA (language) being explored.
- Virtualization may cause as little as 6% loss of cpu.
- Data preservation - plan for this from the beginning.
- Scientific Linux very popular everywhere but JLab.