I'm trying to write a program that will detect signs of failure for portable flash memory devices (thumb drives, etc).

I have seen tools in the past that are able to detect failing sectors and other kinds of trouble on conventional mechanical hard drives, but I fear that flash memory does not have the same kind of predictable low-level access to the hardware due to the internal workings of the storage. Things like wear-leveling and other block-remapping techniques (to skip over 'dead' sectors?) lead me to believe that determining if a flash drive is failing will be difficult at best, if not impossible (short of having constant read failures and device unmounts).

Flash drives at their end-of-life should be easy to detect (constant CRC discrepancies during reads and all-out failure). But what about drives that might be failing early? Are there any tell-tale signs like slower throughput speeds that might indicate a flash drive is going to fail much sooner than normal?

Along the lines of detecting potentially bad blocks, I had considered attempting random reads/writes to a file close to or exactly the size of the entire volume, but even then is it possible that the drive might report sizes under its maximum capacity to account for 'dead' blocks?

In short, is there any way to circumvent or at least detect (algorithmically or otherwise) the use of block-remapping or other life extension techniques for flash memory?

Let me end this question by expressing my uncertainty as to whether or not this belongs on serverfault.com . This is definitely a hardware-related question, but I also desire a software solution - preferably one that I can program myself. If this question is misplaced, I will be happy to migrate it to serverfault - but I do need a programming solution. Please let me know if you need clarification :)

Thanks!