High-Throughput Computing to Enhance Intervisibility Analysis High-Throughput Computing (HTC) • Metacomputing: the network is the computation resource • Run many jobs simultaneously on networked processors e.g. for parameter sweeps, sensitivity analyses, etc. • Processors are not dedicated • Middleware harvests spare compute cycles Schematic Coordinator Starts and monitors Condor jobs. Builds CID. ARC/INFO Condor Intervisibility Analysis •The problem Complex applications based on intervisibility are constrained by runtimes. •Single-processor approaches Attempt optimisations to fit the analysis onto one machine. •Multi-Processor Solution with HTC Simplify the analysis by generating visibility viewsheds for all posts in a Digital Elevation Model (DEM), so creating a Complete Intervisibility Database (CID), accessible for many applications. Condor Middleware •Used to manage concurrent ARC/INFO visibility analyses on networked PCs •Distributes input data and gathers results •Options for CPU-time harvesting: control job start-up and suspension (when other activity increases on a PC ) •Now supports Windows as well as UNIX http://www.cs.wisc.edu/condor/ M.J. Mineter1, S. Dowers1, D.R. Caldwell2 and B.M. Gittings1 1: Geography, School of GeoSciences University of Edinburgh, Drummond Street, Edinburgh EH8 9XP, Scotland Tel +44 131 650 2662 Fax +44 131 650 2524 Email: {mjm, sd, bruce}@geo.ed.ac.uk 2: U.S. Army Engineer Research and Development Center Topographic Engineering Center 7701 Telegraph Road, Alexandria, VA, 22315-3864, USA Tel +1 703-428-6802 X2236 Fax +1 703-428-6425 Email: Douglas.R.Caldwell@erdc.usace.army.mil Prepared for GeoComputation 2003, 8-10 September 2003 DEM Zip Condor Networked PCs running Windows CID Intermediate Files Coordinating processor Design Issues and Results •Computation time 60 CPU hours harvested in 3.5 hours using 37 personal computers •Number of Viewsheds (336*466) dataset: 156576 viewsheds •CID File structure •16 viewsheds are held in one BIL file, for a 4x4 subgrid of observer posts. •Row of BIL files zipped into one file •Enables easy random access to viewsheds •Data volume: Number of bits= (336*466)2. ~ 2.9GB Compressed to 3% or 84 Mb by ZIP •Interactive applications Java used to access the CID; extensions to ArcMap using VBA, ArcObjects are loosely coupled to the Java via shell commands and pipes •Metrics Products descriptive of viewshed characteristics. See poster by Caldwell et al.. Shaded Relief Image Prototype DEM Data Set Southwest Harbor, ME Source: USGS 336 columns x 466 rows