12.07.2015 Views

GSC Sentinel-2 PDGS STBD - emits - ESA

GSC Sentinel-2 PDGS STBD - emits - ESA

GSC Sentinel-2 PDGS STBD - emits - ESA

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>GSC</strong> <strong>Sentinel</strong>-2 <strong>PDGS</strong> <strong>STBD</strong>Issue 1 Revision 2 (draft) - 25.07.2010GMES-GSEG-EOPG-TN-09-0031page 19 of 60These values have been used for the estimation of the needed processing resources withinthe operational system budget analysis.Production workflows on average datastripFront-End Processing 0.72Estimated Time (h)w/o parallelisationup to the Level-0 (consolidated) 1.67Level-0 (consolidated) -> Level-1B without refinement(including 1A and 1B compression)Level-0 (consolidated) -> Level-1B with refinement(including 1A and 1B compression)Level-0 (consolidated)-> Level-1C without refinement(including 1A, 1B and 1C compression)Level-0 (consolidated)-> Level-1C with refinement(including 1A, 1B and 1C compression)Full bulk-Processing: Level-0 -> Level-1C with refinement(including 1A, 1B and 1C compression)35.045.191.2101.2102.2Table 4: Benchmark processing times for selected production workflows (excludingoptional processing steps).3.3 Processing Parallelisation AnalysisTo meet the product timeliness requirements (cf. section 2.2), the <strong>PDGS</strong> processing systemwill implement parallelisation mechanisms such that end-to-end performance can be scaledwith the processing hardware. While some specific steps of the processing algorithm willhave sequential prerequisites preventing parallelisation, most of the others will allow thealgorithm to proceed independently in parallel over multiplied hardware units.This section reviews the parallelisation opportunities inherent to the <strong>PDGS</strong> Level-0 andLevel-1 processing flows which shall lead, when adequately exploited, to a fully scalableend-to-end processing system.3.3.1 PARALLELISATION APPROACHAssuming a large number of available hardware elements, the overall time required forprocessing will be minimised by maximising the number of IPF components run in parallelonto separate elements (e.g. one CPU core).<strong>ESA</strong> UNCLASSIFIED – For Official Use© <strong>ESA</strong>The copyright of this document is the property of <strong>ESA</strong>. It is supplied in confidence and shall not be reproduced, copied orcommunicated to any third party without written permission from <strong>ESA</strong>.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!