Topic: [4+ EH] Slush Pool (slushpool.com); Overt AsicBoost; World First Mining Pool - page 828. (Read 4382671 times)

Quote from: digital on April 22, 2013, 09:37:37 AM

I know slush has an ongoing project related to switching over to DGM. He has been working on it for quite a while now (in his spare time, lol), but I'm not sure how close he is to completing it.

I'm not saying that there aren't any good suggestions out there, but I'm guessing he will just finish that project before implementing any other changes to the scoring algorithm. And in my opinion, that would probably be the best for everyone. We could say goodbye to loosing 25% on rounds under 10 minutes...

Thanks for sharing this. For the reference of others readers, a description of Double Geometric Method can be found here: https://bitcointalksearch.org/topic/double-geometric-method-hopping-proof-low-variance-reward-system-39497
I enjoyed reading it. Note points on periodic rescaling and logarithmic scale.

Cheers,
T

Hippie Tech

legendary

Activity: 1624

Merit: 1001

All cryptos are FIAT digital currency. Do not use.

Quote from: Kruncha on April 22, 2013, 09:29:42 AM

Quote from: organofcorti on April 22, 2013, 09:23:24 AM

Quote from: joolzg on April 22, 2013, 10:25:53 AM

I am happy with the score based calculation as well, re-introducing PPS would not be a good idea. Supporting multiple payout plans would just add unnecessary complexity and demand a lot of effort.

Having said this, I do have some thoughts on the periodic resets:

I could not find the periodic reset documented at https://bitcointalksearch.org/topic/m.50002 which is the "official description" of how rewards are calculated AFAIK.
Theoretically, the exponential nature of the scoring algorithm without the resets would do a great job against hopping.
In practice, for long running rounds, the scores might get very large, causing an arithmetic overflow. Maybe this is the only reason for the resets, but I am just guessing. If so, a mathematically correct fade out instead of the reset might be a viable option.

I might be missing something, so just correct me in case...

Cheers,
T

The resets make no difference to the score. Your proportion is still the same. The only difference the resets make is that they mean Slush doesn't have to deal with numbers with 100 digits.

That's not true, if the score resets on a slow miner and they don't submit a share before the end, they loose everything.

K.

That's called variance. That slow miner could theoretically grab another share or two between the reset and the round ending and end up with a higher score than normal too. You never know when the shares are gonna come. You just can't say that you always loose out, because it's simply not the case.

+1 Timing/ luck is everything.

Last month I decided to have some fun and set the 6870 and 5850 set to diff-42 and the 7870 to diff-105 shares. My then 1070 mhash/s at diff-1, would have averaged 0.005 - 0.008 at the time if I'm not mistaken.

Kruncha

sr. member

Activity: 644

Merit: 250

Quote from: joolzg on April 22, 2013, 10:25:53 AM

I know slush has an ongoing project related to switching over to DGM. He has been working on it for quite a while now (in his spare time, lol), but I'm not sure how close he is to completing it.

I'm not saying that there aren't any good suggestions out there, but I'm guessing he will just finish that project before implementing any other changes to the scoring algorithm. And in my opinion, that would probably be the best for everyone. We could say goodbye to loosing 25% on rounds under 10 minutes...

why, 10 minute rounds are good for everyone, so losing 25% is not good. I would like 24 hours of just 10 minute blocks, but to lose 25% of my reward is not what i would like.

Is there some rational behind this?

joolz

A block is worth roughly (25BTC at time of writing) and as long as you stick with Slush, you'll get the best rewards, because he pays for success, and the more people mining with him means more success.

K.

digital

hero member

Activity: 490

Merit: 500

And don't get me wrong, everyone loves short blocks.

It would just be even better if they met the average payout. And that's why slush has been looking into the DGM scoring algorithm.

digital

hero member

Activity: 490

Merit: 500

I know slush has an ongoing project related to switching over to DGM. He has been working on it for quite a while now (in his spare time, lol), but I'm not sure how close he is to completing it.

I'm not saying that there aren't any good suggestions out there, but I'm guessing he will just finish that project before implementing any other changes to the scoring algorithm. And in my opinion, that would probably be the best for everyone. We could say goodbye to loosing 25% on rounds under 10 minutes...

why, 10 minute rounds are good for everyone, so losing 25% is not good. I would like 24 hours of just 10 minute blocks, but to lose 25% of my reward is not what i would like.

Is there some rational behind this?

joolz

That's how pool hopping works. The pool hoppers will come in for the first x minutes of a round. If a bock is found they profit, if it's not found after x minutes, they leave and go somewhere else, or mine solo.

It's a simplified description of pool hopping, but it results in a lower payout on short blocks for the miners that are here all day every day. If you go down the list of payouts, you will see a pattern where blocks under 10 minutes will have a lower than average payout. This is the reason for that.

joolzg

member

Activity: 76

Merit: 10

Quote from: organofcorti on April 22, 2013, 10:18:18 AM

I know slush has an ongoing project related to switching over to DGM. He has been working on it for quite a while now (in his spare time, lol), but I'm not sure how close he is to completing it.

I'm not saying that there aren't any good suggestions out there, but I'm guessing he will just finish that project before implementing any other changes to the scoring algorithm. And in my opinion, that would probably be the best for everyone. We could say goodbye to loosing 25% on rounds under 10 minutes...

why, 10 minute rounds are good for everyone, so losing 25% is not good. I would like 24 hours of just 10 minute blocks, but to lose 25% of my reward is not what i would like.

Is there some rational behind this?

joolz

Kruncha

sr. member

Activity: 644

Merit: 250

Quote from: Kruncha on April 22, 2013, 10:12:35 AM

Quote from: organofcorti on April 22, 2013, 10:01:58 AM

Quote from: organofcorti on April 22, 2013, 09:23:24 AM

Quote from: organofcorti on April 22, 2013, 10:01:58 AM

I am happy with the score based calculation as well, re-introducing PPS would not be a good idea. Supporting multiple payout plans would just add unnecessary complexity and demand a lot of effort.

Having said this, I do have some thoughts on the periodic resets:

I could not find the periodic reset documented at https://bitcointalksearch.org/topic/m.50002 which is the "official description" of how rewards are calculated AFAIK.
Theoretically, the exponential nature of the scoring algorithm without the resets would do a great job against hopping.
In practice, for long running rounds, the scores might get very large, causing an arithmetic overflow. Maybe this is the only reason for the resets, but I am just guessing. If so, a mathematically correct fade out instead of the reset might be a viable option.

I might be missing something, so just correct me in case...

Cheers,
T

The resets make no difference to the score. Your proportion is still the same. The only difference the resets make is that they mean Slush doesn't have to deal with numbers with 100 digits.

Well, they do make a difference, as it had been pointed out before. My initial thought was to perform the following instead of the reset:

1) Divide by a big number & round everyones score to maintain the proportions but get rid of a lot of digits.
2) Adjust the constant C (currently always 300) according to the big number used for the division - logarithmic. (so at the end, the algorithm maintains the same curve, scaled down)

I do not want to raise a flame war, and note that I was not complaining either - but am open to any discussion on the maths behind this.

Cheers,
T

I'm not quite sure what is it you think the reset does. Can you explain your thoughts on it, preferably with a link to Slush saying something similar?

Basically, it means when the score resets to zero, any slow miner that doesn't submit a share, looses all their shares value (this is to prevent pool hopping, I get that).

If i have 300 shares and the score resets, my shares are only worth anything if i get another one share in before the end of the round.

K.

Unless Slush has changed the way the reset works, that is an incorrect description.

I'm running just 100mhs, Slush hasn't changed anything, I'm not trying to be shitty, it's just how it is. Hopefully Slush can confirm.

I'm not here to cause greif, I want to help...

K.

TiborB

member

Activity: 83

Merit: 10

Quote from: Kruncha on April 22, 2013, 10:12:35 AM

Well, they do make a difference, as it had been pointed out before. My initial thought was to perform the following instead of the reset:

1) Divide by a big number & round everyones score to maintain the proportions but get rid of a lot of digits.
2) Adjust the constant C (currently always 300) according to the big number used for the division - logarithmic. (so at the end, the algorithm maintains the same curve, scaled down)

I do not want to raise a flame war, and note that I was not complaining either - but am open to any discussion on the maths behind this.

Cheers,
T

I'm not quite sure what is it you think the reset does. Can you explain your thoughts on it, preferably with a link to Slush saying something similar?

Unfortunately I did not see this reset documented when Slush explained the scoring calculation (to which I did include the link). We observed that score resets are happening periodically, I assume the reason is to prevent an arithmetic overflow.

The cheapest way (in terms of both db performance and implementation effort) of performing a reset is zeroing out all the scores and simply carrying on the calculation without touching C or scaling back the round time (both of which would be the same). This would obviously not maintain the same characteristics, especially, if the round ends very close to the reset. What I have written is 'scaling down' rather then resetting existing scores, I hope the difference is clear - to maintain the proportions of the scores while also getting rid of digits.

Again, I am drawing conclusions based on observations since I could not find info about how resets are performed - the conclusions might therefore be right or wrong, and I try to be correct/careful about the wording as well. Does anyone know more details of how exactly the scores are currently reset?

Cheers,
T

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: organofcorti on April 22, 2013, 10:01:58 AM

Quote from: organofcorti on April 22, 2013, 09:23:24 AM

Quote from: organofcorti on April 22, 2013, 10:01:58 AM

I am happy with the score based calculation as well, re-introducing PPS would not be a good idea. Supporting multiple payout plans would just add unnecessary complexity and demand a lot of effort.

Having said this, I do have some thoughts on the periodic resets:

I could not find the periodic reset documented at https://bitcointalksearch.org/topic/m.50002 which is the "official description" of how rewards are calculated AFAIK.
Theoretically, the exponential nature of the scoring algorithm without the resets would do a great job against hopping.
In practice, for long running rounds, the scores might get very large, causing an arithmetic overflow. Maybe this is the only reason for the resets, but I am just guessing. If so, a mathematically correct fade out instead of the reset might be a viable option.

I might be missing something, so just correct me in case...

Cheers,
T

The resets make no difference to the score. Your proportion is still the same. The only difference the resets make is that they mean Slush doesn't have to deal with numbers with 100 digits.

Well, they do make a difference, as it had been pointed out before. My initial thought was to perform the following instead of the reset:

1) Divide by a big number & round everyones score to maintain the proportions but get rid of a lot of digits.
2) Adjust the constant C (currently always 300) according to the big number used for the division - logarithmic. (so at the end, the algorithm maintains the same curve, scaled down)

I do not want to raise a flame war, and note that I was not complaining either - but am open to any discussion on the maths behind this.

Cheers,
T

I'm not quite sure what is it you think the reset does. Can you explain your thoughts on it, preferably with a link to Slush saying something similar?

Basically, it means when the score resets to zero, any slow miner that doesn't submit a share, looses all their shares value (this is to prevent pool hopping, I get that).

If i have 300 shares and the score resets, my shares are only worth anything if i get another one share in before the end of the round.

K.

Unless Slush has changed the way the reset works, that is an incorrect description.

Kruncha

sr. member

Activity: 644

Merit: 250

Quote from: organofcorti on April 22, 2013, 09:23:24 AM

I am happy with the score based calculation as well, re-introducing PPS would not be a good idea. Supporting multiple payout plans would just add unnecessary complexity and demand a lot of effort.

Having said this, I do have some thoughts on the periodic resets:

I could not find the periodic reset documented at https://bitcointalksearch.org/topic/m.50002 which is the "official description" of how rewards are calculated AFAIK.
Theoretically, the exponential nature of the scoring algorithm without the resets would do a great job against hopping.
In practice, for long running rounds, the scores might get very large, causing an arithmetic overflow. Maybe this is the only reason for the resets, but I am just guessing. If so, a mathematically correct fade out instead of the reset might be a viable option.

I might be missing something, so just correct me in case...

Cheers,
T

The resets make no difference to the score. Your proportion is still the same. The only difference the resets make is that they mean Slush doesn't have to deal with numbers with 100 digits.

Well, they do make a difference, as it had been pointed out before. My initial thought was to perform the following instead of the reset:

1) Divide by a big number & round everyones score to maintain the proportions but get rid of a lot of digits.
2) Adjust the constant C (currently always 300) according to the big number used for the division - logarithmic. (so at the end, the algorithm maintains the same curve, scaled down)

I do not want to raise a flame war, and note that I was not complaining either - but am open to any discussion on the maths behind this.

Cheers,
T

I'm not quite sure what is it you think the reset does. Can you explain your thoughts on it, preferably with a link to Slush saying something similar?

Basically, it means when the score resets to zero, any slow miners that don't submit a share, loose all their shares value (this is to prevent pool hopping, I get that).

If i have 300 shares and the score resets, my shares are only worth anything if i get another one share in before the end of the round.

K.

digital

hero member

Activity: 490

Merit: 500

I know slush has an ongoing project related to switching over to DGM. He has been working on it for quite a while now (in his spare time, lol), but I'm not sure how close he is to completing it.

I'm not saying that there aren't any good suggestions out there, but I'm guessing he will just finish that project before implementing any other changes to the scoring algorithm. And in my opinion, that would probably be the best for everyone. We could say goodbye to loosing 25% on rounds under 10 minutes...

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: organofcorti on April 22, 2013, 09:23:24 AM