I am experimenting with using the zcu-102 board to import the data in BRAM (2 MBytes) via DMA to the Petal inux of the PS and write it to the SSD as a file.
The contents of the experiment are 2Mbyte of BRAM, and interrupt is generated every 40msec, and the data is fetched from the PS through DMA.
I'm trying to save 50Mbytes per second, but I do not see any performance.
If I open and close the file without writing it to the file, I see performance. But I write it to a file, I get 32Mbytes per second.
Perhaps the file is written to the kernel and seems to be slowed by buffer caching.
Is there a solution ???
I tried both O_SYNC and O_DIRECT as flags when opening a file, but performance is not improved.