[1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration!

[ciuciu]

Inaba,

After a lot of pool testing, at the moment I'm looking at PPS or SMPPS.
  
I gave your pool again a chance (I mined with you previously, until your pool went down for a full night), because I really liked your site and style.
All people are screaming pool hopper, but you and me know that this is not the case. I might return in case you implement one of the two methods.

[1984]

Inaba,

After a lot of pool testing, at the moment I'm looking at PPS or SMPPS.
  
I gave your pool again a chance (I mined with you previously, until your pool went down for a full night), because I really liked your site and style.
All people are screaming pool hopper, but you and me know that this is not the case. I might return in case you implement one of the two methods.

If you have 15Gh of miners then I'd suggest at least having one fail-over pool, most pools even the larger ones suffer from outages for various reasons from time to time. Having the much mining power relying on one pool doesn't seem like a good idea to me, with that much power I'd be running a mining proxy to help with logistics to point all your miners at it and let it deal with pool selection. Also if you're mining 24/7 you'll struggle to beat the scoring method used on eclipse unless you hop.
Just my 0.02bc, thanks for the extra bonus btw.

[Inaba]

I don't think anyone is saying you are a pool hopper.  I have not seen any previous behavior to indicate you were hopping.  I'm sorry that you got unlucky on the last block.  I have had similar things happen to myself - but as someone already said, I don't get or keep the coins.  They are equally and fairly distributed between all the miners; there's no benefit to me if someone gets unlucky.

Hopefully I will have the double geometric method worked out soon, which should ease some of that burden.  I worked on it this afternoon, but was having trouble integrating it with the current tables.  I am going to start over will all new tables and work from there this week. 

[ciuciu]

1984, I'm using cgminer.
I just wanted to try again Eclipse, as I like to give the smaller guy a chance, but it seems Arsbitcoin is a better fit for me. Good luck!

Inaba, you didn't say it, but all the other implied it. I will check your pool after you implement your new method.

Regards.

[Meni Rosenfeld]

Inaba, you didn't say it, but all the other implied it.

This is getting ridiculous. Are you even reading what I'm saying?

There is no attempt to detect hoppers, accuse hoppers, or penalize hoppers. I don't think you're hopping, you don't have anything to gain by hopping this pool, and I or participants in this pool couldn't care less if you're hopping since they wouldn't lose anything from it.

All there is is a scoring method which is mathematically proven to reward everyone fairly no matter if they're mining continuously, intermittently, hopping, or whatever. It has the unfortunate side-effect of having high variance, thus you could be unlucky and get much less than the expected reward, or lucky and get much more. This averages out over the long run. With the double geometric method it will be both hopping-proof and have much less variance.

[ciuciu]

OK, I get it but I do not care.
It is a stupid method, even if mathematical makes sense. How do you expect me to mine on this pool, when my first reward instead of 4 is 0,004? Should I go on hoping in 1 year will even out? This is a bullshit method and stupid variance, and sooner or later you will lose most users.

The end.

[kripz]

1.42325800
(+20.08%)

I got 20% more because you and others disconnected for a lengthy period of time before the block was solved The pool operator gets nothing (though inaba is mining so he gets his share).

Unless you find a pool that is straight PPS or prop then you are going to have high variance. With prop you will lose BTC to hoppers.

Not in the middle, it says it right there in the original post of the thread:

Not everybody comes here and reads the forum before signing up. I tell my friends hey, visit eclipsemc.com and sign up. They have never even been on the bitcoin forum.

[organofcorti]

stupid variance

It's the only thing you've written that I agree with, and I and many other miners have said it before. Stupid variance!

Stupid variance is stupid.

[ciuciu]

I'm speaking about the variance introduced by the scoring method.

[Inaba]

I will make the scoring method more prominent on sign up or on the main page so people are aware of it.  I'm going to really dig into the double geometric method tonight and see if I can get a working version going.

[bacon strips]

I am liking the geometric method so far and need to read about the double geometric method when I get the time.  My current payout is 30 Bitcoins and I have a 5.62% above proportional pay.  I mine 24/7 for the most part so I am happy with the extra 5.62% over proportional. 

[WebMonkey]

i would prefer double secret geometric but i'm fine with whatever.

i mine 24/7 so none of it really bothers me one way or the other.

'monkey

[Inaba]

SShhhhh!

[Inaba]

I have made decent progress on the double geometric code.  Assuming it's approved by Meni as being valid, I will move on to coding the payout and estimate code tomorrow night.  Currently, I have the scoring code in place and counting on a test server with a couple of my miners to make sure it's doing it's thing properly.

[Inaba]

The double geometric code is on the live servers and it is populating the new tables.  I missed the start of the new block, so we'll see how the numbers play out for this block.  If it looks good, I may make it live the next block or the block after if something is amiss.

We need to come up with a description of the system that is easy for people to understand and also explains why the system is superior to proportional and many other systems in a language that is as simplified as can be.  I figure I'd through that out there and see if anyone can distill it better than I can.

[cengique]

*sigh*

I don't mean to annoy you, just telling you the practical experience from the point of view of the miner. I'm glad now, instead of PPLNS, we're considering the double geometric method, which I assume is similar to the current method. Is there a post explaining it?

Especially in the long blocks, people hop-off to leave and mine on other pools.

They don't gain anything from doing that. The fact that the round was long does not affect the payouts of futures shares they submit.

I agree with you. I didn't claim they have a gain from hopping off the pool. All I said is that people get tired when the block gets long and decide to mine on other pools. Maybe it's not the only reason, but I often see the total hash rate fall from 80 to 60GH/s over the course of a long block. In the last monster block, I remember we fell to about 50 GH/s. Then it becomes important that the scoring method makes people stay in the pool. If you didn't have that I assume more people would be leaving. I don't have a way of proving this, though.

If they can guarantee payment by contributing in the last N shares without any decay, they can do this without loss.

They don't know when the round will end, so they don't know which shares are the last N. There's decay, but it's a step function rather than exponential.

Ok, I wasn't completely correct, there is decay because as there are new shares mined, the window slides and your shares will go down. However, consider the following disadvantages of using the PPLNS method:

1) In the first N shares of a block, PPLNS is equivalent to PPS.
In the mineco.in example, N was 750k shares, and with Eclipse's hash rate, it would be reached in about 17 hours. So if a block is shorter than that, it would practically be found using PPS method.

Inaba: Could you add the average values at the bottom of the block stats table? It would ne nice to know our actual average shares/block and also their standard deviation maybe.

2) After the pool has mined N shares for a block, during the time it takes to mine N shares by the pool, one can hop in for n << N hours and would have (N-n) hours without any decay in reward.
They do not gain anything from doing this, but the pool's hash rate is reduced. If they switch to a giant pool, they get steady payout.

3) In a long block, if many people follow the idea in (2), the pool's hash rate can drop to zero, stopping the rewards from decaying and the pool never succeeding in finding the block.
Obviously this is an extreme case, but if the hash rate drops significantly it will create a similar scenario where the pool becomes "cursed".

The geometric method, in addition to being hopping-proof, also encourages the miners to stay put in the pool.

No, it does not. For the past shares they will get the same reward whether they stay or quit. For future shares they will get the same reward whether they mined previously or not.

I think there's a misconception that decay only happens when you leave the pool. But past shares decay the same way whether you're in or out. The reward for future shares is independent.

Thanks for the explanation, it does help to understand the probabilities better. You are right, except in my case (2) above, where there will be a short time without decay for the past shares. Although, I don't challenge the fact that your reward will increase in both methods if you mine continuously.

[Meni Rosenfeld]

I'm glad now, instead of PPLNS, we're considering the double geometric method, which I assume is similar to the current method. Is there a post explaining it?

Sure, but get ready for disappointment. It is a hybrid between PPLNS and the geometric method, and I recommended to Inaba to set it to be close to PPLNS, to reduce the variance. I still don't get the problem with PPLNS. But this is exponential PPLNS rather than 0-1 PPLNS, so your objections may not apply.

Especially in the long blocks, people hop-off to leave and mine on other pools.

They don't gain anything from doing that. The fact that the round was long does not affect the payouts of futures shares they submit.

I agree with you. I didn't claim they have a gain from hopping off the pool. All I said is that people get tired when the block gets long and decide to mine on other pools. Maybe it's not the only reason, but I often see the total hash rate fall from 80 to 60GH/s over the course of a long block. In the last monster block, I remember we fell to about 50 GH/s. Then it becomes important that the scoring method makes people stay in the pool. If you didn't have that I assume more people would be leaving. I don't have a way of proving this, though.

Well. If there's a need, I can design a method that really does encourage people to stay. But, not being a magician, I cannot design a method that encourages people to stay and gives people a reason to start mining in the first place. And, I think any method which is not 100% hopping-proof (expected payout etc. per share always the same) is ignoble, and that the problem you describe is negligible.

1) In the first N shares of a block, PPLNS is equivalent to PPS.
In the mineco.in example, N was 750k shares, and with Eclipse's hash rate, it would be reached in about 17 hours. So if a block is shorter than that, it would practically be found using PPS method.

I don't follow this logic at all. You are paid once per share in PPS. In PPLNS, you can be paid for every share 0,1,2 or any number of times.

2) After the pool has mined N shares for a block, during the time it takes to mine N shares by the pool, one can hop in for n << N hours and would have (N-n) hours without any decay in reward.
They do not gain anything from doing this, but the pool's hash rate is reduced. If they switch to a giant pool, they get steady payout.

They could have started mining in the giant, low-variance pool in the first place. If they started mining for you and then leave, you can't force them to stay.

You may be interested to know that unlike PPLNS but like the geometric method, double geometric can reduce pool-based variance (the variance caused by the pool being too small). This is at the cost of increased operator risk, though, and I don't know how risk-loving Inaba is.

3) In a long block, if many people follow the idea in (2), the pool's hash rate can drop to zero, stopping the rewards from decaying and the pool never succeeding in finding the block.
Obviously this is an extreme case, but if the hash rate drops significantly it will create a similar scenario where the pool becomes "cursed".

I don't see that realistically happening.

[Reuef]

I am getting close to 50% rejected shares using Phoenix 1.6.2 with RollNTime support.  Have switched for now will try again after this block is solved.

[Inaba]

A word about the new Double Geomtric Scoring (DGS) system:

EMC will be moving from the Geometric Scoring system to the Double Geomtric Scoring system.  Twice the GEOMETRIES!  ZOMG!!11!1one!!

But seriously the new system will be somewhat different than the old system, and perhaps a bit unfamiliar to some people, so want to go over some of the major points and answer any questions people might have.

First off, this system is a hybrid between the current geometric system and the PPLNS system.  In our current system, your shares geometrically decay, becoming worthless after a period of time (in our case, about 5 hours) - if you stop mining, after ~5 hours you will no longer receive a reward.  This has the advantage of making the pool hopper proof, but the disadvantage that there would be a chance you would not be paid due to your shares decaying.

With the DGS, your shares decay much more gradually, over not only the number of shares but also the number of blocks. What this means is that there is effectively a sliding window of time where you will be paid for shares submitted.  Each share submitted is paid a score which will eventually be converted into BTC but over time (shares AND blocks), the score decays.  This happens at a much slower rate than with the geometric system, thus reducing variance on a per share basis and spreading your score decay over several blocks.  A consistent miner will eventually reach a point where your hashrate is in proportion to the pool and your payout will be the same as an unhopped proportional pool on a per block basis.  If you are inconsistent, your payout will also be inconsistent - however, you will always eventually be paid for the work you've done regardless of when or how consistently you have submitted that work, until your score decays past the point of value.

To be a little more technical, your score is an estimated average pending reward.  When a block is found, some of your score is converted into BTC up to the total amount of BTC generated (50 BTC basically), once everyone has been paid evenly up to 50 BTC amount, the rest of your score is stored going into the next round and the process will happen again.  On really long rounds, this means that the pool will pay out a bit of extra from it's own pocket to pay off everyone who contributed and on short rounds the BTC will be stored for future unluck rounds.

This method is completely hopping proof, so there will be no losses to pool hoppers.  You can jump in and out of the pool and expect to be paid fairly for what you've contributed, like a PPS pool.  The variance is much lower than the geometric method and similar to proportional while the maturity time is much shorter than the *SMPPS methods.  Because the risk to the pool is greatly reduced, we also are not forced to charge exorbitant fees to hedge future losses on long rounds.

For a more technical explanation of the system, you can see this thread: https://bitcointalk.org/index.php?topic=39497

I welcome any questions or comments you all have in regards to this new system.  I plan on making it live within the next block or two, after I have verified that it is paying out properly vs the current system. Currently it is running in tandem with the geometric scoring so that I can verify the output and payout.

[kripz]

Still techinical imo, hybrid between pps and pplns? what's this?, If thats for the website you will have to make it more simple and treat them like they are completely new, assume they dont know any other payment method and just sell your own.

ran it by my friend and he actually tried but concluded with tl;dr.