Low Memory Profile (LMP) Hash

[WhoSoup]

Hey, wanted to get your take on this idea:

It is possible to have an lxr hash that does not load the bytemap into memory, at the cost of sacrificing almost all performance. The idea is to just have a file pointer open to the bytemap file and use Seek to jump to the appropriate offset, then read a single byte.

This is very inefficient compared to keeping it in RAM so this would only be useful in situations where a low RAM usage is critical. Not suitable for mining.

Regular LXR was ~151ms for 1000 hashes
With an older SATA ssd, LMP was ~7400ms for 1000 hashes
With a newer M.2 ssd, LMP was ~3400ms for 1000 hashes

Anyway, this is probably an unreasonably poor utilization of file i/o that it's not a viable option, just wanted to put it out there.

[Emyrk]

I think it's a valid option to offer.

[PaulSnow]

I'd note that if you really only want to validate a few blocks of hashes, then this is a viable option, and then trusting the history going backwards (avoiding the hashes) would work just fine for some wallets and infrastructure.