Topic: [1050 TH] BitMinter.com [1% PPLNS,Pays TxFees +MergedMining,Stratum,GBT,vardiff] - page 230. (Read 837101 times)

This is not what he wrote about. The coins received are not the coins from the withholded blocks but the coins from the usual payout (for share submission).

This is patently false. Please think about what you are saying before you start typing such things.

There is no way to perform a MITM attack or block withholding attack where the miner withholding receives the coins instead of the pool.

I do pool audits for US$120 (1.29 btc). PM me if you'r really interesting in knowing for sure if there's a problem or not.

It may have sounded counterintuitive because redtwitz was assuming a normal distribution in a situation where that assumption can't be made without significant loss of accuracy.


If a PPLNS pool has lower fees than a PPS pool, then eventually then probability that you earn more on a PPLNS pool will approach 1.0 over time.

Thanks for your post. It helps me clear things a bit between probability and the normal distribution function. I guess that the fact that luck "is not stateful over time" is counterintuitive to me. Fortunately I'm not gambling ;-)

However, I think PPLNS supporters aren't always very clear about the fact that there is a risk of earning less EVEN IN THE LONG RUN than pure PPS. That's the part that I had missed. Even if one risks just as much losing than winning, the risk is still here compared to PPS. Or am I wrong again?

Shares per round / D is erlang distributed, so the lower tail CDF for the data you provided (100 @ 1.216005475, 250 @ 1.0990548393, 500 @ 1.0534418012) is 0.9800301, 0.9379993, and 0.8828175 - or 2.00%,  6.20% and 11.72%. So for example, in 100 repeats of 100 blocks, you'd expected to get an average that large or greater 2% of the time. As you say, hardly unusual.

There will never be more than 50% probability of being lucky, just as there's never more than 50% of probability of being unlucky.

Past luck doesn't affect future luck. It doesn't matter if the pool was lucky or unlucky recently; the probability of finding a block after n times difficulty shares is always the same.

While there's certainly a possibility of evening out the past bad luck or even surpassing the expected rewards in the long haul, this cannot be the likely (p > 0.5) scenario.


Luck is just that: luck. There's no explanation for it. You either have bad luck or you don't.

Taking the last 100, 250 and 500 blocks from the statistics, I found that the mean of the required shares to solve a block was 1.216005475, 1.0990548393 and 1.0534418012 times difficulty, respectively.

Assuming a normal distribution and estimating the standard error from the respective samples (0.1262058917, 0.0744550916 and 0.0485402741 respectively), there's a chance of 4.36%, 9.18% and 13.57%, respectively, that a given pool would be that unlucky.

4.36% sounds pretty low, but since there are more than 23 mining pools, chances are that it happens to one of them...

Need some figures if I'm going to help - number of blocks in the time period, the number of shares per block, and difficulty of each block. Or I'll do it tonight when I get home.

Here are some figures that I get from the 2nd graph on the rewards page, and from the table on the shifts page.

As I write this, the graph starts at shift 14348, that was on 2013-04-17 11:05 i.e. 15 days ago.

The pool has never quite been lucky since then.

Starting at shift 14670, 2013-04-27 04:23 i.e. about a week ago, the pool has started running unlucky, more than I have ever seen in the past (but I have only joined about 2 months ago). That is why I asked the question about variance and expected "recovery" time before the actual rewards catch up with the expected ones.

I think that the theoretical, correct answer to the question "how long before we catch up" is: you cannot know when, you can only know that it *will* happen in the future. So maybe I should rephrase the question:

From our current "unlucky" status, how long would it take before we have statistically more than 50% probability to catch up with the expected rewards? I believe this question can be (formally) answered. I'm not really expecting it, but if someone is both able and willing to calculate or even approximate it, I would find it useful and appreciate it :-)

On a separate topic ... any advantages of using BitMinter's Javascript client miner, versus BFGMiner on stratum?

Indeed - that's the graph I was pointing at (the one at the top is too confusing and doesn't really allow any meaningful interpretation) - that bottom graph is only last 500 shifts so maybe last 14-15 days worth, and I agree that right now the luck of the pool is bad.  Remember even if the pool is 10% more unlucky than average at the moment, you still get the +9% from namecoin, and over time (longer than two weeks) the luck all evens out.

Will

You'll have to ask other pools why they don't. There is nothing in the Stratum protocol to make merged mining any more difficult than it is with getwork and getblocktemplate. When I added Stratum and getblocktemplate I just didn't turn off merged mining.

Not all pools have their own software. Some run one mining server software they found on the net 2 years ago that supported getwork and merged mining, but isn't maintained anymore. Then they found a new one that does Stratum but not merged mining, and they installed that too. The result is a patchwork of unmaintained outdated software and a random feature set. Other pools do have their own software and just left out merged mining because the mergeable coins looked dead. That was a pretty good assessment too, until someone suddenly started buying very large amounts of namecoins, pushing the price higher than it had ever been before.


New work is pushed on Stratum when there is a new BTC or NMC block. New work is also pushed every 30 seconds to include new transactions.

That does not imply a restart, however, as there is no progress when mining. If you lost the lottery every week for 10 years it doesn't mean that your chances of winning are now magically higher.


I didn't have the impression we have been plagued with bad luck for months. I don't see that when I look at organofcorti's weekly pool stats.

And no, changing the data to be hashed does not cause bad luck. That's what your Stratum miner does when it creates new work anyway, it creates new data to be hashed. It doesn't even matter whether you change the entire merkle root or just 1 bit in the timestamp to create new work. Changing just 1 bit in the data being hashed will cause many changes in the resulting hash. This is called the avalanche effect.

Let me put it this way: if you can find a way to generate new work in a way that affects your success rate when mining, then you have found a loop hole in SHA-256 and you'll be a rich miner.

I'm not saying it's impossible that we randomly bumped into a weakness in SHA-256. But I don't think it's likely. It's not time to call Bruce Schneier just yet.  

Entropy, is that true, has it been some months of lower than expected payout?  I was wondering why bitminter can do merged mining but slush can't on stratum...

We had 22 blocks In a day about 2x normal.  That is about best I can remember . The best thing to do is go on this site.


https://bitclockers.com/calc 

wait for the difficulty to change it did on tue this week so  put in your hash and see what you should get 10 days in a row. My hash is 9Gh/s today I should earn .44919 coins. ship your coins off and set to 0.  then set your coins at 10x or in my case 4.4919  . 

 check every day  same time.


 after 24 hours your should be at .44919
 after 48 hours you should be at .89839   

 once the ten days are up see how close you are to the calculator..  then multiply by 1.1 why the nmc you earn are worth 10% more.  so if you earned 4.3 coins in 10 days and said damn I am short multiply by 1.1 and you are at 4.73 coins and that my friend would be a good 10 days. most pools charge more then bitminter.

i do advise you to put some money in asicminer if you can't get an asic you can get a share for about 1.2 btc   

most of my money is in gpu farms...... 7k
  then some in bfl .......................... 1.6k
then some in asicminer shares.............  0.5k  total of 9k invested . 

I have collected about 11k in coins.

This page only shows the last 500 shifts. Payouts have sometimes been much higher than average. I have on my TODO list to make it possible to go back in time on that graph, but I haven't had time to add it yet.

If you scroll down that page further, a second chart shows cumulative payout. Seems like there is a huge variance in terms of expected avg payout and actual (4940 vs 4463 at the time of this post). That's nearly 10% below average!

Also note that the payout is hardly ever above average, and the rare times it is, does not appear to be 10% above.

Go to https://bitminter.com/stats/rewards and then scroll down and look at the graph at the bottom.

If the green line is below the yellow line, then the pool is currently unlucky.  If the green line is above the yellow line, then the pool is currently lucky.

Will

Could anyone with more PPLNS experience than me tell us about how unlucky the pool has been over the last few days? Is this usual variance or not, and should we expect the "actual" rewards graph line to cross the "expected" line anytime soon?

I know about the theory behind PPLNS and why it should compensate naturally over time, I'm just trying to match theory and practice here

Thanks for your feedback.

Ah, I forgot about prop. And of course there's prop score, too. Why do you think people mine at Slush's pool? They can pronounce the reward method.

How about an indicator on the side of the chart that would show the average of the dots? Or an average line?