OTB pipelining in memory or disk

OTB applications can be chained on disk or in memory. To evaluate the gain in performance, the 2 ways has been compared.

A python code chaining SARCalibration and Orthorectification has been written and run on HAL.

ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=4

• On disk RAM=256Mb

         real    2m57.092s
         user    6m3.192s

• In memory RAM=256Mb

         real    3m48.434s
         user    9m16.246s

• On disk RAM=4096Mb

         real    2m53.787s
         user    5m35.072s

• In memory RAM=4096Mb

         real    1m50.522s
         user    2m56.159s

For information, here are the time for each OTBApp:

SARCalibration

• RAM=256Mb:

          real    1m28.333s
          user    4m28.946s

• RAM=4096Mb

          real    1m25.232s
          user    4m36.138s

Orthorectification

• RAM=256Mb

          real    1m19.415s
          user    1m9.509s

• RAM=4096Mb

          real    1m14.660s
          user    1m2.282s

The image are on /work filesystem. I didn't try to use the node local disk ($TMPDIR)