Does anyone notice that their miners generally runs better until the first flushwork?
I get around 1% HW errors, and after a flushwork occurs, it shoots up to 3-4%. The HW ticker goes from 1-7 a jump to upwards of 50 a jump.
YES, I noticed that all along... I think the flush-work....still needs work!
****
It seems to "Flush" with every single block detected, and not for just the pool you are on, which causes the errror rate to be higher than it has to be...drastically.
IMHO...that's the biggest tweak needed atm
You are way off on this one...
Flushwork has to run with every block detected on the network. Blocks are built on top of each other network wide, not just on your pool.
Hmm, I still respectfully disagree. Here's why...
I never had that problem mining with GPU's... and it never flushed anything other than our own "Stale" shares.
In pool mining...We don't solve blocks for other pools, we work on our own blocks, and by running flush-work every time a block is detected, you literally loose every workshare you are currently solving, even though your pool hasn't found it's block yet.
My pants are literally loose.
But no, all miners (and pools) are working to add a new block to the end of THE block chain. Pools don't each work on their own separate chain.
Each time a new block is added to the chain, all miners (and pools) stop their current work, and begin new work, trying to extend the now longer chain.
The block trying to be added is unique to the pool, but the chain it is being added to is not.
(There are rare exceptions involving hostile miners, but let's not complicate the discussion with those cases.)
hmm, ok...
Try this.... look/watch the error rate jump @ flushwork times.
Given the jumps, why would the dumped/flushed shares contribute to errors if the pool didn't want them?
You will also notice the lack of a jump in errors during the flushwork when your pool is finding the block.
I hear what you are saying, but the errors I monitored over the course of mining say otherwise.
I can point my Raedon card at the same pool, on a separate account, and physically see the difference.
Flushwork on it has no bearing on error rates on the card, and certainly shouldn't on a knc machine either, but it does, which proofs a glitch in flushwork imho
I'd rather get rejected shares at the appropriate time than have it flushed at every single block sniffed on the chain, or at least try it.
The machines also solve shares quicker than flushwork is performed it looks to me.This needs tested... no flushwork at all /vs/ flushwork
The higher the diff your shares are, the more relevant it becomes
This is all irrelevant really if your pool pays for stale shares.... every flushed share is a loss!
