I'm not saying there isn't a problem, but this was his response a while back, and I don't know if he made it before you edited your post to say "nvm it took a while but went up to 2."  I think it was originally set to 20, and I looked for the post where he mentioned the changed numbers, but couldn't find it.

Website is down?

Machines are still hashing on EMC

Inaba, did PayPal ever give you back your money when they froze everything? Just curious.

Depends on what you define as a problem. It hovers mostly around diff 1 and occasionally rises to diff 2 at 37 shares per minute. But you said your target share rate is 30pm for US1 which I thought you said originally was slated for 20pm.

Ok, wondering whether you used any hysteresis for diff up versus diff down and whether the diff target is a "keep below 20 shares". There's nothing actually wrong here then, just my misinterpretation.

Sorry to hear that.. hope they give it back one day. Doesn't seem right at all.

Ah, got it... no, it just uses a sliding window.  Is there some drawback to bouncing around?  The work is accepted for old difficulty work, so it doesn't matter if there is a difficulty change, since the difficulty is tied to the work sent out.  You might be currently working on Difficulty 3 and sending back difficulty 2 shares from an older set of work that hasn't been sent back yet.

EMC tracks difficulty per job, not per user.  So it works on both the getwork and GBT side, so it doesn't matter which protocol you use or when you submit the work, as long as it's not stale, it will accept it if it's under the difficulty for a particular job.  If you're discarding work when you receive new work without submitting it then yes you'd lose shares, but if you submit it even after you get new work, it's likely still valid within the stale window.

Woo - nice.
That's my argument against Stratum at the moment, the fact that difficulty is separated from work.

Now if we could just get GBT to reduce the amount of data that it needs to send to the miner (more than GetWork with roll-n-time), use a better protocol ..., and solve the mystery of ignored transactions; then if Stratum's diff was also fixed we'd have 2 closely competitive options ...