I am using SNAP version 8 under Window Srv 2019, for Speckle Filtering operation.
Due to the server using 4x CPU (16 core per CPU) for running this program, we try to use all the CPU physical cores for the operation, but there will be partial CPU been used.
I have tried modify the snap.parallelism parameter in snap.properties, but it did not work.
number of CPU cores used for image rendering and graph processing
allow this to default to Runtime.getRuntime().availableProcessors()
snap.parallelism = 72
And the “Number of Threads” value in “Performance” setting only can detect 48, is it a limitation of SNAP setting?
or Is there any way to run all the CPUs ?
Thanks.
Many remote sensing workflows are dominated by I/O (reading/writing files). Getting the best performance from a NUMA system is complicated. Often, using fewer cores will be significantly faster than using all cores due to better utilization of CPU caches, and there are ways to control which CPU’s are use for which tasks that can help with cache utilization. Each task will have a different optimal configuration, but it takes a lot of work to find the optimal configuration for a given task. If you will be doing the same processing many times, it is worth trying different settings. On one workflow using a 24-core linux system with hyperthreading but very slow storage, using more than 8 CPU’s increased the run time.