ADASS was held in Paris, 6-10 Nov.
Fermi attendees: TonyJ, Johann, Jurgen and me.
First day had GPU and R tutorials; I attended GPU. Nice intro to them going as far as snippets of code and strategies for porting code systems. They say all the tutorial info will be online.
Starting to see GPU clusters: 100-500 GPUs from 1-3M$. Claim is massively parallel computing paradigm is the future, even if GPUs fade away. Intel introducing the MIC (Many Integrated Core) chip with 32 cores as potential competitor.
Display using GPU — GLnemo2 It uses Qt, openGL with rendering on GPU. Knows several filetypes, including Gadget (Cosmic ray propagation code).
Clouds: high level discussion. I/O remains expensive; cycles are competitive. There are not guarantees on data integrity — it can get lost.
HPC in Astro tidbits: Look up Pegasus workflow management, DagMan workflow engine, Condor Schedd scheduler.
FutureGrid being set up as free academic cloud.
GPUs and python: pyCUDA for compiling and linking CUDA code into python. Can also use the python C-API for host-based code.
Knowledge Discovery: Discussion of algorithms used by Google et al: self organising maps, pattern searching. They plan to run this latter one on Fermi unids. Google Clusterings Labels Scores Patterns Spotter.
VAO tools: see http://www.usvao.org/tools/. Some are SAMP enabled.
R in astro: growing exponentially in user contributions. Interfaces to python.
http://www.r-project.org
http://cran.r-project.org/web/views
SciDB competitor: monetdb, SciQL
SAMP: messaging for applications and web apps. Like ds9, topcat, aladin, WWT.
http://ivoa.net/samp/
Richard Dubois
15 November, 2011