working on a PC with 8 GB RAM and Windows 10 with the latest version of SNAP 5 and the latest version of SNAP 6 Preview I got the error messages
- GC overhead limit exceeded
- Cannot construct DataBuffer
when I tried to calibrate (complex values or at least 2 output bands) data of a subswath of a SLC product in different configurations of SNAP (e.g. heap size of JVM of 3 GB up to 5.7 GB).
The problem apparently depends on the size of input or output product, e.g. calibration of sigma0 alone works.
In the task manager of Windows one can see that about 1 minute before an error pops up the cache-disk is 100% busy while the data disk is idle and memory consumption is between 96% and 100%.
There are several reports about this problem in the forum but I have not found a solution except more RAM for the PC.
My first workaround was the use of a Google compute engine with Linux and 26 GB RAM. This worked just fine.
Then I have done the same operations in a virtual machine (VirtualBox, 4,5 GB RAM) with Debian 9.2.1 (Stretch) and the actual SNAP 5 linux-version on my PC. It worked (but of course quite slow).
That’s puzzling and I would like to know:
What’s the cause of this difference of SNAP/Windows and SNAP/Linux?
Is there something like a configuration switch to enable the (successfull) memory management of SNAP/Linux for “small RAM” in SNAP/Windows?
Clarifying comments are welcome. Thanks in advance.