Quote from: organofcorti on December 24, 2012, 08:06:20 AM

Quote from: Luke-Jr on December 24, 2012, 07:28:53 AM

Quote from: organofcorti on December 24, 2012, 07:06:17 AM

Quote from: Luke-Jr on December 24, 2012, 06:52:20 AM

The background you've posted is wrong. Eligius has always paid out immediately (or as close to it as possible); while there were some delays in SMPPS due to the variance of block finding times, the maximum rewards ever took was a mere 3 days since I made sure to send before any balance was waiting that long.

Well, it was nice of you to dip into your own pocket to pay miners, that's not a part of the SMPPS protocol.

Obviously these payouts came from the pool wallet, which gets filled by the SMPPS buffering and when miners haven't earned enough to achieve a reasonable payout yet.

But the pool wallet had insufficient funds to pay extra credit - isn't that why there was a back log?

There was a backlog because the pool was rewarding miners with buffer. This was mostly before we dipped into extra credit - Eligius went about a year(?) straight lucky. At no point was raw (ie, not converted to reward) extra credit paid of course.

Quote from: organofcorti on December 24, 2012, 08:06:20 AM

Quote from: Luke-Jr on December 24, 2012, 07:28:53 AM

Quote from: organofcorti on December 24, 2012, 07:06:17 AM

Quote from: Luke-Jr on December 24, 2012, 06:52:20 AM

Your blog misrepresents extra credit as "owed", which it is not and never has been.

This is something you've mentioned many times, and I don't understand it. If extra credit is not owed, why pay it? If you didn't pay it, would miners still earn 100% of PPS?

It's paid as part of the reward system rules, as an incentive for miners to keep mining even on long rounds when the buffer is empty. No reward system can pay 100% PPS, with or without extra credit. SMPPS and CPPSRB just do the best they can to try to achieve that.

Do you mean irl or in theory? Theoretically, any provable fair reward method without a fee can pay 100% PPS. If you mean irl, then ok, but the extra credit is an integral part of SMPPS and if the extra credit isn't guaranteed to be paid even in theory, then it even theoretically it may not pay 100% PPS. I understand this never happened for SMPPS, but I would like to understand your thoughts on the matter better.

Theoretically, reward systems can pay 100% PPS given infinite time. SMPPS's extra credit is part of the mechanism it uses to achieve this; many other systems use variance to do it. In other words, extra credit is only as obliged in SMPPS, as if proportional pools are obliged to have lucky rounds. In theory, both (EC rewarding, and prop lucky rounds) should happen - but there is no guarantee/debt of them.

Quote from: organofcorti on December 24, 2012, 08:06:20 AM

Quote from: Luke-Jr on December 24, 2012, 07:28:53 AM

If the extra credit in SMPPS was ignored/didn't exist (PPPS), it would underpay miners during long rounds after it went broke, just like proportional always does for long rounds.

This is why I thought discriminating between a "buffer" and "extra credit" was wrong. The point of extra credit is to enable fair rewards, and to achieve PPS.

The discrimination is necessary and important. The buffer is guaranteed (the pool already has it!), but nothing can guarantee future findings - no matter whether it's SMPPS's EC, proportional's lucky blocks, or any other system.

Quote from: organofcorti on December 24, 2012, 08:06:20 AM

Quote from: Luke-Jr on December 24, 2012, 07:28:53 AM

Quote from: organofcorti on December 24, 2012, 07:06:17 AM

Quote from: Luke-Jr on December 24, 2012, 06:52:20 AM

CPPSRB also tracks a value similar to "extra credit", but the same still applies. With CPPSRB, extra credit is actually less likely to ever be paid due to how LIFO works. On the upside, CPPSRB gives actual rewards much more like proportional, so they fit with the actual block payout availability much better without any manual sends; in normal operation, no payout should be delayed more than a single block, and most (at least on Eligius) will be paid out in the same block they are earned in.

I read the info on the Eligius site for the first time, and it's quite an interesting reward method. By assuming that 100% of PPS value won't be paid (am I right in that?) you're able to make certain that reward are paid in a timely manner. So time to maturity is reduced to 100 network block confirmations, with a trade off of increased earning variance. I think most miners would be happy earning a little less if they didn't have to be worried about when it may get paid.

Have you derived (or otherwise calculated) the expected PPS miners will earn under the new method?

With Eligius's no-fee implementation, shares are still paid 100% PPS value, but the unavoidable loss is in the backlog of shares that inevitably never get paid. I think it's easiest to explain if you compare it to PPLNS; the main differences there are just that shares are never paid twice (which results in older shares getting paid on lucky rounds). While PPLNS effectively forces a pool to compromise between variance (low N, like <=expected) and reward delays (high N, like expected* Cool

, CPPSRB's rewarding older shares rather than doubling rewards gives it a low variance with instant rewards.

Nice analogy - that makes it clearer, thanks. However it seems that there is a possibility that in the long term not all shares would be paid.

In practice, it is pretty much guaranteed that some shares will go unpaid ever. That end result is, again, unavoidable no matter what reward system is used.

Quote from: organofcorti on December 24, 2012, 08:06:20 AM

However, informed miners are happier miners. Including, for example, how many shares on average won't be paid in a given time period - a month in your explanation may help. Or the probability that any given share won't be paid in a given time period. It may also be possible that some shares will never be paid - and if so miners should know the probability of that, too - if it's possible to derive.

The probability a share is never paid is pretty much identical to the probability of blocks being stale. Considering all the factors involved there, it can vary quite a bit (as much as 5% at extremes, it seems) and isn't likely something that can be predicted.

Quote from: organofcorti on December 24, 2012, 08:06:20 AM

It may be that very few shares won't be paid long term and have no significant impact on miners, but I think it would be good for miners to know what that might be - even if it's just the result of a simulation rather than a derivation. Have you managed anything like that?

I don't know any simulation of stale blocks. I'm not sure it's even possible.