BitMinter audit (or: PPLNS vs PPS)

[DrHaribo]

Recent bad luck at BitMinter had some miners worried, or even moving to pools with high fees, no transaction fee income, and no namecoin income, because they felt it would be more profitable than the bad luck at BitMinter, which they thought was so bad it must be more than bad luck, a bug of some sort.

I commissioned organofcorti to do a detailed analysis based on hard facts:

http://organofcorti.blogspot.com/2013/05/131-bitminter-and-luck.html

We did indeed have bad luck lately, but take note of this: our payouts beat most pools even when we have bad luck! And even without counting the income from namecoins, which is almost 8% extra income at the moment!

So if you are paying high fees (3+ %) for PPS mining in a pool that doesn't even pay out transaction fee income or namecoins, then you really have to ask yourself:

Instead of sometimes high and sometimes low pay, did you choose to get extra-low pay consistently every day because it feels safer? Yes, you are guaranteed to never have a decent payout ever. You are safe from that. Although you did get rid of the element of luck, I don't think this is really what you wanted.

Thanks to organofcorti for a very detailed analysis. If you need to make sense out of numbers, he's the man for the job.

TLDR; when we have bad luck, PPS miners are sad. When we have good luck, PPS miners cry. 

[1713906440]

1713906440

[1713906440]

1713906440

[1713906440]

1713906440

[1713906440]

1713906440

[1713906440]

1713906440

[juhakall]

Now, if you just added DVC & IXC merged mining, Bitparking miners couldn't claim that they are earning more than Bitminter miners.

Of course, throwing 12TH/s at those chains would disrupt the status quo a bit.

[organofcorti]

Big influx of new miners who haven't gotten their heads around the whole pooled mining thing. I wonder which pool will be next to have some miners freak out on them?

[DrHaribo]

Is there any place to sell ixcoins now? Vircurex wrote on Twitter that they won't do IXC trading anymore. They are apparently down right now due to being hacked, though.

I may consider adding DVC, if there will still be an exchange supporting them.

[juhakall]

Vircurex was the only place to trade those coins AFAIK. IXC was added back after the initial decision to remove it, but maybe only the removal was announced on twitter?

[shmadz]

Just wanted to drop in and say thank you to OOC for offering the audit (even though you were paid by the pool operator, I don't think anyone can question your integrity*)

and also a thank you to Dr. Haribo. The simplicity of your miner client is unmatched. I can't wait to test it with an ASIC! (if I ever get one... 4 months and counting )



*edit: that bit about questioning OOC's integrity is badly worded, especially in a bitcoin world. I love bitcoin because I shouldn't have to trust anyone. everything is verifiable through the math. I think I must have chosen that phrase because I personally don't have the time or ability to actually check the math myself. What I should have said is that I, personally, have no reason to question OOC's integrity.

[juhakall]

Vircurex was the only place to trade those coins AFAIK. IXC was added back after the initial decision to remove it, but maybe only the removal was announced on twitter?

Vircurex is now up and trading continues with both DVC & IXC. DrHaribo, please save us profit-seekers from DGM! I would much rather mine on a PPLNS pool.

[organofcorti]

Just wanted to drop in and say thank you to OOC for offering the audit (even though you were paid by the pool operator, I don't think anyone can question your integrity*)

and also a thank you to Dr. Haribo. The simplicity of your miner client is unmatched. I can't wait to test it with an ASIC! (if I ever get one... 4 months and counting )



*edit: that bit about questioning OOC's integrity is badly worded, especially in a bitcoin world. I love bitcoin because I shouldn't have to trust anyone. everything is verifiable through the math. I think I must have chosen that phrase because I personally don't have the time or ability to actually check the math myself. What I should have said is that I, personally, have no reason to question OOC's integrity.

Ah, I knew what you meant - anyone can verify my integrity

[eleuthria]

Ah, I knew what you meant - anyone can verify my integrity

Shill!  Shenanigans!  Fluffed audit!  Clearly paid propaganda!  We're onto you!   /s

[organofcorti]

Ah, I knew what you meant - anyone can verify my integrity

Shill!  Shenanigans!  Fluffed audit!  Clearly paid propaganda!  We're onto you!   /s

But can you say that with 95% confidence?

[eleuthria]

Ah, I knew what you meant - anyone can verify my integrity

Shill!  Shenanigans!  Fluffed audit!  Clearly paid propaganda!  We're onto you!   /s

But can you say that with 95% confidence?

I'm not even confident of what I ate for lunch

[redtwitz]

According to this forum topic, Vircurex doesn't have an official Twitter account. The last tweet is from April 7, so even if the information posted there was accurate once, it's now outdated.

Vircurex is back up. Their website says:

We currently support Bitcoin, Namecoin, Devcoin, Litecoin, Ixcoin, PPCoin, Terracoin, Feathercoin

[Pionwave]

Ah, I knew what you meant - anyone can verify my integrity

Shill!  Shenanigans!  Fluffed audit!  Clearly paid propaganda!  We're onto you!   /s

I heard it was a reptilian conspiracy...

Switching from BTCGuild to Bitminter PPLNS about a month and a half ago, without using any of my big pile'o'nmc, i'm still doing just as well.

I'm new to Bitminter, and one thing that wasn't clear and i think would be great in the statistics is the amount of historical blocks solves daily showing the trend. As there are some rough days.. but over the month and half it has been fairly even.. adding the trend may help the newer miners to see that it's worth slugging through the rough days.

[organofcorti]

Ah, I knew what you meant - anyone can verify my integrity

Shill!  Shenanigans!  Fluffed audit!  Clearly paid propaganda!  We're onto you!   /s

I heard it was a reptilian conspiracy...

Damn Reptoids!

Switching from BTCGuild to Bitminter PPLNS about a month and a half ago, without using any of my big pile'o'nmc, i'm still doing just as well.

I'm new to Bitminter, and one thing that wasn't clear and i think would be great in the statistics is the amount of historical blocks solves daily showing the trend. As there are some rough days.. but over the month and half it has been fairly even.. adding the trend may help the newer miners to see that it's worth slugging through the rough days.

The thing is that there should never be a trend in data consisting of independent identically distributed random variables (iid). Looking for a trend would only serve to test if the data was iid or not, and that's not easy or intuitive for miners to follow - it would just be a p-value of a particular test to find trends.

Where people often see trends is in moving averages. The problem with that is the moving average can't be said to have a trend, since the data it contains is no longer iid - each moving average data point is influenced by previous data points. Same goes for cumulative means and sums.

As an example, if you take data that is by definition random ( say generated random data) and take a moving average of the variables, you'll also see "trends". However since we already know the data is random, there can't be any trends.

That's not to say a moving average has no use - we can use it to test when data is no longer within certain confidence intervals, and might help guide us to data that appears "wrong". But the use to which many miners put a moving average is to predict future luck, and that can't happen. I'm paraphrasing what someone here wrote, (eleuthria or DrH, or maybe gyverlb? Can't remember) but we can't speak of periods of good luck or bad luck. Luck has happened in the past, but that has no bearing at all on luck in the future.

This is a problem that we've seen before in p2Pool - even someone as mathematically literate as kano fell into the trap of considering a moving average as a trend, and I didn't hadn't figured out how to explain why it couldn't.

Edit: Another example - if you take the same data the shows a "trend" and instead of a moving average of n periods, just take the average every n periods, there is no detectable trend at all.

tl'dr: Trend analysis is hard, and moving averages can't show trends.

[eleuthria]

organofcorti, possibly easy question, or possibly extremely difficult to answer:  Due to the random nature of block solves, is it accurate to say that on average, a pool will spend more time with negative "luck" than with positive "luck"?  I want to say that's the case, and it's why miners are always pointing out how <X> pool has consistent bad luck.

In my eyes:  Bad luck is DEFINED as a long period with fewer blocks than expected, while good luck is defined as a lot of blocks in a shorter period.  So you would expect that a pool would probably spend most it's time negative when viewed on a graph with time as the X-axis.  However, the good luck bursts only need to last a fraction of that time to "make up" for the bad times.  When viewed on a graph, that would paint the picture of consistent bad luck, but when evaluated over a long enough period, you would find it gets very close to the expected average.


I'd love some confirmation/observations from you on the matter [maybe split to another thread, but it does fit in with this BitMinter thread as well, so it's your choice].

[organofcorti]

organofcorti, possibly easy question, or possibly extremely difficult to answer:  Due to the random nature of block solves, is it accurate to say that on average, a pool will spend more time with negative "luck" than with positive "luck"?  I want to say that's the case, and it's why miners are always pointing out how <X> pool has consistent bad luck.

In my eyes:  Bad luck is DEFINED as a long period with fewer blocks than expected, while good luck is defined as a lot of blocks in a shorter period.  So you would expect that a pool would probably spend most it's time negative when viewed on a graph with time as the X-axis.  However, the good luck bursts only need to last a fraction of that time to "make up" for the bad times.  When viewed on a graph, that would paint the picture of consistent bad luck, but when evaluated over a long enough period, you would find it gets very close to the expected average.

I'd love some confirmation/observations from you on the matter [...]

This is a really important question, because along with interpreting moving averages as trends, reporting shares per round / difficulty wrt to time is incorrect and as you say leads to incorrect interpretations of the data. What "luck" is depends on how you define it. If you take a time average of shares per round over difficulty, it will be on average worse than expected. A moving average will appear to spend more time indicating bad "luck" than good.

I always show shares per round / difficulty "luck" measurements either wrt to number of blocks solved by the pool (for example moving average measurements) or grouped (for boxplots). Even so, in the Bitminter piece I still did boxplots wrt to a difficulty period when I really should have done it wrt to number of blocks solved - but in that case I was trying to show how variance increases when the number of solved blocks is reduced.

You could still chart a moving average of shares per round / difficulty wrt to blocks solved and just add the dates corresponding to the block number on the x axis.

Another possibility that in some circumstances would be unaffected by a daily or moving average is blocks per day. This will be a Poisson distributed random variable if the pool's hashrate never changes, and so would be possible to average over time without distorting the perception of "luck" - as long as you provide confidence intervals as well.

However, pools do not maintain a constant hashrate wrt to difficulty, so instead you'd have to calculate  blocks per N shares (think PPLNS) which is only Poisson distributed wrt to the submitted shares, but not to time. So again, you can't average this type of "luck" measurement by day either.

The way I get around this in the weekly pools stats is to provide a CDF probability for the number of shares it took to solve the number of blocks during the week. Then we expect to get a bunch of random dots each week. If the dots don't look randomly distributed, then there's a problem. Unfortunately this doesn't seem intuitively understandable to most miners. I like BitMinters  CDF based "luck" page, and it's easy for me to see at a glance what the pool's luck has been, but it seems most miners can't.

Maybe the easiest thing to do would be changing the charts to show luck wrt to blocks solved and add in the corresponding dates.

[maybe split to another thread, but it does fit in with this BitMinter thread as well, so it's your choice].

A new thread's a good idea - people can ask questions and I can try to answer them, or look for answers if I can't. Or maybe other people can help answer. I'll set up a thread when I get back from Mother's day dinner.

[Pionwave]

tl'dr: ...moving averages can't show trends.

You're right, didn't think that through.. wishful thinking I suppose.

[candoo]

@DrHabiro please add DVC & IXC.  :>

[Bonam]

@DrHabiro please add DVC & IXC.  :>

Seconded. More merged coins = more profit for everyone on bitminter. So why not?

[candoo]

@DrHabiro please add DVC & IXC.  :>

Seconded. More merged coins = more profit for everyone on bitminter. So why not?

Yeah thats the question. Why not? There is nothing to loose