Max

I did notice the node fell behind due to IO bottleneck. I did not see the "waiting for farmer to accept ..." messages though.

I limited the maximum parallel re-commit threads to 32 and it went through without crashing. The node also didn't fall behind during the rehash this time. So my guess the...

Dirty hack, a global counter barrier in the worker threads.

SSD is probably fast enough to not have the node fall behind. I tested on 24 disk (10k, 2.5 inch) RAID6, with the node and farmer data both on that...

Maybe just test a normal setup, and hammer the drive with some benchmarks to make the node fall behind but not wait on farmer.

Then it may depend on the large plot size after all.

Btw it happened 3 times out of 3 for me (on same machine), so it's not random.

`_UINT128_T_STATIC` allows linking of DLLs which themselves have uint128_t statically linked, if I remember correctly.

Looks like it's just slow, https://github.com/madMAx43v3r/mmx-node/releases/tag/v0.8.9 should be better

not sure what you mean by parsed?