Clarification on the SSD usage for pmdk

[charles...@storagecraft.com]

From the glossary link at https://pmem.io/glossary/

Since the programming model for persistent memory is based on memory-mapped files, the PMDK libraries will work correctly on top of any NVM technology (like an SSD with a traditional file system on it).

So, the assumption would be the pmempool create would just be on an file in the filesystem onto of the SSD. This would fallback to the msync calls since it wasn't a DAX device or anything special and all of the associated performance issues in this case.

Would there be any other setup or expectations needed to test/use the PMDK onto of an SSD?

Thanks in advance. 

[Andy Rudoff]

Hi Charles,

No, no additional setup is required.  You can use PMDK on any file on any disk (i.e. on your laptop) to do development and the libraries will work as expected with one very important exception: the libraries will not perform very well.  The algorithms are designed for pmem and tuned for pmem, so when you don't actually have persistent memory, the algorithms in PMDK will still produce correct results, but the performance is not anything you'd want to ship in a product.  We use regular files all the time when developing & testing code, but then we always switch to actual pmem for benchmarking and final validation.

Hope that helps,

-andy