Eligius POLL: Reward system changes, and new ASIC-ready Eligius-Hu pool

[Luke-Jr]

People seem to be against HU (according to the polls)

But nobody's willing to say why; makes me wonder if it's just the trolls who want Eligius dead.

Or people just don't want to gamble on an untested payment strategy?

PPLNS is untested? Really?

Now SMPPS, when we originally switched to it, THAT was untested.

[imsaguy]

People seem to be against HU (according to the polls)

But nobody's willing to say why; makes me wonder if it's just the trolls who want Eligius dead.

Or people just don't want to gamble on an untested payment strategy?

PPLNS is untested? Really?

Now SMPPS, when we originally switched to it, THAT was untested.

How many pools use PPLNS as you have proposed, using the values you've proposed?

[Luke-Jr]

People seem to be against HU (according to the polls)

But nobody's willing to say why; makes me wonder if it's just the trolls who want Eligius dead.

Or people just don't want to gamble on an untested payment strategy?

PPLNS is untested? Really?

Now SMPPS, when we originally switched to it, THAT was untested.

How many pools use PPLNS as you have proposed, using the values you've proposed?

PPLNS isn't DGM. The value doesn't change how it works.

[eldentyrell]

How many pools use PPLNS as you have proposed

p2pool and 5 other pools according to https://en.bitcoin.it/wiki/Comparison_of_mining_pools

, using the values you've proposed?

Is there something wrong with tuning the lone parameter of this algorithm?  That's a far cry from a whole new payout system, which is what you were making it sound like (and which it is not).

[eldentyrell]

At this point, Eligius is no longer the only pool that allows a miner to mine anonymously with just a bitcoin address - other than the stats graphs, I'm not sure what is unique about it now that may have been previously.

X-Roll-Nonce.

Works wonders for my cell phone bill.

[torusJKL]

PPLNS is ok if new blocks are found frequently.
With another miner I had the case that no block was found for a long time, then my network stopped working (weekend). Until the network started functioning again I was passed the N value, the block has been found and I got nothing from it.
That is why I got back to PPS.

[organofcorti]

I hope this goes ahead as is, but if it's just N = 8 that has people skittish, then it's just a discussion about a preferable N.

At N = 8, it will take (at current hashrate) about 3 days for each share to be paid in full - this is the cost of the minimal variance. Since Eligius miners are used to much longer periods of time to payout it shouldn't seem like a downside at all.

PPLNS is ok if new blocks are found frequently.
With another miner I had the case that no block was found for a long time, then my network stopped working (weekend). Until the network started functioning again I was passed the N value, the block has been found and I got nothing from it.
That is why I got back to PPS.

The expected value of any share under PPLNS is the same as that share's value as a result of a PPS reward. The difference is variance, and variance at this pool under N = 8 will be very low. I think (but I'm not sure) that this would make it suitable for intermittent mining.

Congrats on taking this huge leap, Luke-Jr. It's not the easiest thing to do, changing a reward method that has become associated in chief with you and your pool, but your miners will thank you for it.

[racerguy]

PPLNS is ok if new blocks are found frequently.
With another miner I had the case that no block was found for a long time, then my network stopped working (weekend). Until the network started functioning again I was passed the N value, the block has been found and I got nothing from it.
That is why I got back to PPS.

that problem wouldn't have happened where n=8x difficulty.

[kuzetsa]

People seem to be against HU (according to the polls)

But nobody's willing to say why; makes me wonder if it's just the trolls who want Eligius dead.

Shame on you for saying "trolls" and "wanting eligius dead" in the same breath as you quoted me.

1) I voted a "support" option for the eligius-RA system (effectively, a new system which sounds very much like wizkid's proposal which was sought-after by myself, and several other voters)
2) double shame on you, because I already asked plenty of questions on IRC, and more than one person in this thread has expressed specific concerns.
3) here, have some reasons that I myself don't like Eligius-HU one bit.

For starters, Eligius-HU is largely undocumented, and not explained well in the OP.

What is the size of a share anyway? How many hashes is that?

I ask because the entire bitcoin network is usually between 7k and 8k blocks per month, of which, eligius only finds 100 or so... how does this arbitrary value of "8x difficulty" fit in for our rate if we suddenly mine more or less percentage of the bitcoin network's blocks?

The current "second proposal" doesn't really explain the math which will determine the size of the window over which each miner's recent shares are counted for the purpose of payouts.

Also, my original paranoid questions which I suspect others may have had similar concerns about not getting paid for the previously submitted shares under the Eligius-RA PPS system.

This last concern ties in with my first new question -- Eligius-HU system is going to arbitrarily change the size of a share to be dynamic based on hash-rate, so that nobody's math will be able to know what is going on with the new algorithm.

[imsaguy]

People seem to be against HU (according to the polls)

But nobody's willing to say why; makes me wonder if it's just the trolls who want Eligius dead.

Shame on you for saying "trolls" and "wanting eligius dead" in the same breath as you quoted me.

1) I voted a "support" option for the eligius-RA system (effectively, a new system which sounds very much like wizkid's proposal which was sought-after by myself, and several other voters)
2) double shame on you, because I already asked plenty of questions on IRC, and more than one person in this thread has expressed specific concerns.
3) here, have some reasons that I myself don't like Eligius-HU one bit.

For starters, Eligius-HU is largely undocumented, and not explained well in the OP.

What is the size of a share anyway? How many hashes is that?

I ask because the entire bitcoin network is usually between 7k and 8k blocks per month, of which, eligius only finds 100 or so... how does this arbitrary value of "8x difficulty" fit in for our rate if we suddenly mine more or less percentage of the bitcoin network's blocks?

The current "second proposal" doesn't really explain the math which will determine the size of the window over which each miner's recent shares are counted for the purpose of payouts.

Also, my original paranoid questions which I suspect others may have had similar concerns about not getting paid for the previously submitted shares under the Eligius-RA PPS system.

This last concern ties in with my first new question -- Eligius-HU system is going to arbitrarily change the size of a share to be dynamic based on hash-rate, so that nobody's math will be able to know what is going on with the new algorithm.

tl;dr

Basically, Luke-Jr, you did a terrible job explaining things and then wonder why there's a negative response. 

[kuzetsa]

...long post is long...

Basically, Luke-Jr, you did a terrible job explaining things and then wonder why there's a negative response.  

((edit: Can still click "Quote from <<name>>" to view the cited posts / context))

... This summary is slightly inaccurate, as the shares will be of dynamic difficulty depending on your hashrate, and the value of each share will depend on your share-difficulty and the current block difficulty at the time the share is submitted, to avoid being hoppable during difficulty changes, and gracefully scale to ASICs as they are available...

Yes, but only slightly. My CPU mining & GPU (nVidia) rigs are only slightly slower than ASICs. The variance can be disregarded because I deserve to be paid just as much per share. (( this is obviously satire on my part. I really don't understand, and even if Eligius-HU is fully explained, this "dynamic difficulty" will scare me, in addition to the other concerns ))

(( edit 2)) Just noticed, luke, you quoted me out of context:

...the decision has basically already been made.  This is just to justify the decision.

Yeah well it's not working. People seem to be against HU (according to the polls)

What I meant was that the votes don't seem to be justifying the change(s) after the fact (was that really the plan?)

[SleeperUnit]

When I first read up on PPLNS, I didn't think I would like it. The reasoning being that with my limited mining capacity and the fact that this is my main PC and therefore can't be expected to mine full time, I Imagined the unfortunate scenario where I keep submitting shares that don't make the cut and therefore don't get payed out. I understood that other shares would get payed out multiple times and that my earnings should still average out to full PPS value but I just didn't want to face the prospect of some of my work going unpayed.

At that time, I figured that N would have to be rather large for me to stay on such a pool. I decided I would leave if N was less then difficulty and was really hopping it would be closer to 2 x difficulty. With N = 8 x difficulty, every single share returned to Eligius within the last year would have been payed at least once. As far as I can tell, only Eligius-De was ever unlucky enough to have blocks go longer then 8 x difficulty.

Alright so I guess I can live with PPLNS but what about the current system, how's that working out for me? So I just did some calculations based on the last 10 blocks on stats page for my address and here's what I found.

Over the last 10 blocks I got payed at an average rate of 0.00001716 BTC / Share. At current difficulty, the expected pay rate is 0.00002282 BTC / Share.

So the current system is costing me 25% of my expected earnings because the pool had some bad "luck" over the past couple months. Until we can catch up and start a new buffer, no sensible person should mine here because of what is essentially a 25% fee to pay off people that already quit mining here.  Yes, there is still the very remote possibility that the pool's luck will swing around and pay out all that EC but I'm not going to hold my breath.

With PPLNS, past luck has no impact on future payouts, so there is no bad time to join such a pool since "luck" only affects payouts on the previous couple days worth of shares.

[kuzetsa]

When Eligius-RA shuts down, some people will probably want this configuration:

primary configured mining pool: ra.mining.eligius.st
fallback mining pool: something other than eligius.st (HU)

Please do not hijack ra.mining.eligius.st as a redirect to the new HU system

Simply shut the IP address down or whatever, allowing people to gracefully switch. Doing it this way will allow people to use "active mining" variant of backpay on their existing extra-credit balances before you drop everyone like they've gone out of style.

This policy is one I am recommending because the OP didn't specify what would happen when Eligius-Ra ultimately becomes unavailable.

Regardless of this policy recommendation, I'm just going to use my existing, unmodified configuration which points everything at the mining.eligius.st address, because I have no desire to see the pool "die" or whatever.

With Eligius pool(s) since I've joined, I've had a very good sense of community, and I get along well with many of the active participants, contributors and developers (all the scripts and features that make it nice, irrespective of any BTC backlogs or political issues with the selected payout system)

[Luke-Jr]

People seem to be against HU (according to the polls)

But nobody's willing to say why; makes me wonder if it's just the trolls who want Eligius dead.

Shame on you for saying "trolls" and "wanting eligius dead" in the same breath as you quoted me.

I didn't  know you voted "dislike Hu", nor did I meant to imply you were a troll (I wouldn't have responded at all if I thought you were).

For starters, Eligius-HU is largely undocumented, and not explained well in the OP.

What is the size of a share anyway? How many hashes is that?

I ask because the entire bitcoin network is usually between 7k and 8k blocks per month, of which, eligius only finds 100 or so... how does this arbitrary value of "8x difficulty" fit in for our rate if we suddenly mine more or less percentage of the bitcoin network's blocks?

The OP tries to explain it in language everyone can understand. It sounds like you want some low-level technical details here... but those aren't entirely worked out yet.

The current "second proposal" doesn't really explain the math which will determine the size of the window over which each miner's recent shares are counted for the purpose of payouts.

Huh?

This last concern ties in with my first new question -- Eligius-HU system is going to arbitrarily change the size of a share to be dynamic based on hash-rate, so that nobody's math will be able to know what is going on with the new algorithm.

Where's the question?

...the decision has basically already been made.  This is just to justify the decision.

Yeah well it's not working. People seem to be against HU (according to the polls)

What I meant was that the votes don't seem to be justifying the change(s) after the fact (was that really the plan?)

imsaguy's post is an example of a troll I ignored. I figured the troll nature of it should have been obvious considering he said the same nonsense about wizkid057's last poll too.

[kuzetsa]

...  The OP tries to explain it in language everyone can understand. It sounds like you want some low-level technical details here[...] but those aren't entirely worked out yet  ...

Was not asking for your low-level code, nor for any specific reference implementation in any language whatsoever... The technical details could be in a high level language, pseudocode, or even a list of goals and logical assertions about how various system states and cases will interact, and what the intended result will be.

The implication which you just made: "things are going to be changed" (details that still need worked out) between now and the final implementation is the main reason I am against the Eligius-Hu change. I am of the opinion that a system which is not clearly defined or explained... No. Just no.

((edit retracted))

[Comepradz]

I personally agree to HU, but Luke-Jr please, we have a serious business here. Maybe you should make some simulation of HU based on current mining capacity, compare with RA's, forecast for future ASIC miners and show us the resulting data. I cannot say that some people are too paranoid or even xenophobic for new setup, maybe some of us don't agree because it's new or don't know the data. Thank you.

[TheSeven]

So Luke-Jr, you say you want explanations why people vote "dislike HU". I am one of them.

While PPLNS is a well-understood payout system, it has itw own set of flaws, mainly high variance. Even at 8xD. (And counting which shares will have been paid at least once is nonsense, because that's equivalent to merely 12.5% PPS payout.) I dislike variance, and that's why I was at Eligius until things started to go bad (which I predicted months ago, but nobody would listen). I'm very willing to pay a 1-3% cut for true PPS payouts, and also accept modeare payout delays if the pool has bad luck. What I don't accept is share-based variance.

Another problem is that current EC credit will likely go unpaid to a large percentage in the end, which basically screws Eligius' most loyal miners!
(I'm personally mostly unaffected by this though, I've realized what was going to happen early enough.)

Sure, the current situation doesn't look good, and we need a change to fix that. The root cause of all this trouble is that for various reasons (orphan blocks, possibly withholders, ...) a true 100% PPS payout isn't sustainable in the long run by definition, but Eligius' non-equalized SMPPS payout model tries to pay each share 100%, regardless of what is feasible. If it can't, it will start to neglect more recent shares at the benefit of less recent ones, which leads us to where we are today.

So which payout level is sustainable in the long run? 95%? 98%? 99.5%? This depends on a lot of factors, and only time can tell.
What we need is a payout system that takes this fact into account.

I can think of two possible models:

• 95%-CPPSEB
  Assuming we're certain that 95% PPS payout is sustainable in the long run, one option would be a 95% Capped Pay Per Share with Equalized Backpay system.
  This basically means that, if the round was sufficiently short, each share of that round will be paid 95% PPS immediately. If the round was longer, it might be paid less. If the round was shorter, the excess credit is used to increase the PPS percentage of older shares (the ones that were paid the least).
  This increases the chance of immediate 95% payouts for active miners, while ensuring that after some time all shares will have been paid almost equally much, however much is feasible in the long run (this dynamically adjusts itself).
  
  Old EC from the current system could be carried over. If this route is taken, new shares would basically get a 5% fee at first, which is used to pay off old EC up to 95%. Once that point is reached, the payout level of all shares will slowly be increased to whatever is feasible.
  
  See this link for more details on CPPS-based reward systems.
  
  
• ESMPPS(AM)
  Equalized Shared Maximum Pay Per Share is a system that basically works like Eligius' current reward system, just that it prefers the least-paid shares instead of the oldest ones. It will thus first pay the shares of the current round until they have been paid as much as the least paid previous shares. Then the remaining block reward will be used to increase the payout percentage of the least paid shares (including the shares from the current round).
  This will show similar dips to CPPS* or PPLNS (just that for the latter they wouldn't be brought up to speed later) during long rounds, much better than with SMPPS in EC mode. So even if our luck would be down to 50% for a month, all shares would be paid about 50% PPS, and as soon as luck gets better, those shares will be backpaid by more recent higher-luck blocks. In the end this should end up paying about >98% PPS (whatever is sustainable, dynamically adjusting) to all shares after a while. Recent shares might see slight temporary dips, but not worse than with PPLNS, which doesn't even fill those dips later.
  
  Even though this should not be necessary with an ESMPPS system, one could extend it to prefer old shares of currently active miners to keep people loyal. It would just use 90% of the block reward to pay shares of currently active miners (considering at most equally many old shares of a miner as he has mined in the current round, like it was proposed for Ra), up to a payout level of 100% PPS. The remaining 10% would be distributed across all old shares to eventually pay out all EC if luck permits. If not all of the 90% can be allocated to active miners, the remainder will go towards all shares as well.
  
  Old EC from the current system could be carried over in two ways:
  
  • Consider every old share at its current payout state. This means that the system will try to reach equal payout levels on all shares, independent of when they were found (before/after the switch to the new model).
  • Consider every old user/address at its current payout state. This means that the system will at the beginning prefer users that have been hit particularly badly by the current "EC mode sesion". Old users which have been paid an average of 98% PPS during their whole "eligius career" wouldn't be paid any of their EC until all users which had a worse "career" have been brought up to speed. I consider this approach more fair, and it will get the PPS percentage for newly mined shares into the 90+% area much faster than the alternative above.
  
  See this link for more details on ESMPPS.
  
  

One additional advantage of these models is than (unless the AM addition is applied, which I don't consider necessary), these models are ready for a situation where transaction fees become predominant compared to the block reward, as it allows for >100% PPS payouts if transaction fees are included in the rewards, and luck + fees allow for >100% PPS.

Sure, if the switch to one of these models would have been performed much earlier, they would recover from the current situation much easier, but I don't think it is too late yet.

I simulated the ESMPPS model like a year ago when I was cooperating with another pool owner on finding a good reward system. The results looked pretty good. Sadly the pool it was invented for didn't really take off and never had more than like a dozen gigahashes, before it was finally shut down. I'm fairly certain that the reward system wasn't to blame though.

I demand that, before a switch to PPLNS is performed, these two models will at least be simulated and analyzed.

Feel free to contact me with any questions about these two proposed systems, preferably in #eligius on irc.freenode.net.

[Luke-Jr]

I simulated all these reward systems a year ago: http://eligius.st/~luke-jr/samples/800MH-3/

The problem with 95%-* systems is that it has no potential to ever be any better than PPS with a 5% 'fee', but still has the risk of the pool hitting long unlucky streaks. Everything else equal, why would anyone pick 95%-CPPSEB over 95%-InstantPPS?

[donking]

Just guarantee that any outstanding extra credit will be paid, and then implement whatever you want. And tell people when you
have made up your mind what you think is best.

Then people can accept the changes, or go somewhere else.

As long as all the people who have outstanding extra credit gets paid. The problem is if somebody who have been doing work is going to be dealt
the short stick and not get paid in this switch.

[TheSeven]

I simulated all these reward systems a year ago: http://eligius.st/~luke-jr/samples/800MH-3/

Well, that is one sample. But to decide on the feasibility of any reward model, you'll definitely have to take several different situations into account. I usually simulate at least 1000 runs and look at best/worst/average case and their probabilities afterwards.
Psychological questions are also surely a factor, not every miner will decide purely rationally based upon what's expected to result in the best payout in the end. Some prefer higher payouts, some prefer less risk. Some even mine PPS on deepbit.

The problem with 95%-* systems is that it has no potential to ever be any better than PPS with a 5% 'fee', but still has the risk of the pool hitting long unlucky streaks. Everything else equal, why would anyone pick 95%-CPPSEB over 95%-InstantPPS?

Because what I proposed in fact has the possibility of being better. If there is any backpay total payout cap at all (only really needed for -AM), that would still be 100%. What I propose is to cap immediate payouts at 95% and use the remainder for (possibly >95% or even >100%) backpay. This is just to prevent shares from "overshooting" above the long-term sustainable payout rate and increase the amount of backpay to finally increase the chance that all shares meet up at the (~98%) final sustainable payout level.