[1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration! - page 208.

Meni Rosenfeld

donator

Activity: 2058

Merit: 1054

Quote from: Inaba on July 21, 2011, 08:47:00 AM

It's not a trivial thing to implement... incidentally the "cheat proof scoring" system in Simplecoin is all sorts of broken, which was the base I was starting from and that has turned out to be a huge mistake. I should have just written it from scratch like everything else.

This explains a lot. Some time ago someone started a pool and boasted a "cheat proof scoring". Upon asking, it became clear he had no idea what that meant and he just chose the "cheat proof" option from his pool software. I wondered if that was actually my geometric method, and if so, how come it was implemented painlessly after it has proven to be nontrivial to others who tried. So it looks like Simplecoin was the pool software, and that it didn't have a correct implementation after all.

There may be others who try to use it so I think we need to warn them.

... And it looks like I may be to blame for the error in Simplecoin, seeing as I misunderstood something I was asked about.

Quote from: Meni Rosenfeld on June 07, 2011, 10:45:30 AM

Quote from: simplecoin on June 07, 2011, 10:35:48 AM

So, I'm close...... it looks like max is the previous row lscore value, is that right?

Yes.

Inaba

legendary

Activity: 1260

Merit: 1000

It's not a trivial thing to implement... incidentally the "cheat proof scoring" system in Simplecoin is all sorts of broken, which was the base I was starting from and that has turned out to be a huge mistake. I should have just written it from scratch like everything else.

Besides, it's a learning experience for me

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Sorry this is turning out to be so much work for you. Worth it for your members, but clearly a pain in the arse all the same. No wonder so many pools are content to stay proportional.

Inaba

legendary

Activity: 1260

Merit: 1000

Are you proposing that for a proportional or score system (or either?) ...

Scoring numbers are coming back together a bit better, there's still two hiccups to work out and I'm working with Meni to solve them. I ran it against the reference data set and block 24 and they both came up with the same numbers (which is good) but those numbers are wrong (which is bad). Once that computational error is nailed down and there's the difficulty change-mid-block code in place, it will be ready for prime time.

btcboston

member

Activity: 101

Merit: 10

What you are suggesting is sort of a variation of the pay per last N shares system.

PoulGrym

member

Activity: 83

Merit: 10

Just an idea I been thinking of.. If anyone gets me.. please do re-explain this..

What about an hourly avg "Rudder" system.. Where an avg share sending system scans a window of one hour sent in shares and calcs the avg of them, not just the amount but Shares per Minute SPM (BPM).. If you jump off the ship (pool) for one hour.. It will take another hour to get the full speed/direction back.

The Kicker would be.. once a block is found.. It will not shout out it's found until one hour after. Then it will take the avg from one hour before the found block time window and the hour after the found block window avg.. Then the average of those both will be the last avg. This number should be the hopping score + prob of total shares sent in.

Does this make any sense? How can you translate it in to math?

THe higher the pool's hashing speed the more blocks will be found in less time. Still it should be one hour window even when a block might be found with in an hour. Forcing you to stay in the pool.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: PoulGrym on July 20, 2011, 04:02:23 AM

What ever it was you did.. it's working.. Rejected shares have been very low! <1%

Just need to get scoring working.Can't have those hoppers jet as soon it's not worth it anymore.. We need to stick to it.

It's not just the 'jetting as soon as it's not worth it'. It not just a question of increasing variance, it's more a question of non-hoppers losing money, and hoppers coming out around 140% to 200% ahead. That coinage comes from non-hoppers.

Inaba

legendary

Activity: 1260

Merit: 1000

I'm still building the reference dataset for score testing. It's almost done, I should have made the dataset smaller from the start, but since it's so far along, I figure I'll let it finish.

At that point, I'll run the scoring algorithm through it and see what comes out the other end and run it by Meni, assuming he's still willing to generously help out. Once it's confirmed working, I'll put it back into place for this current round or next round depending on how long this one lasts.

PoulGrym

member

Activity: 83

Merit: 10

YAY NEW BLOCK FOUND! Go go!!

btcboston

member

Activity: 101

Merit: 10

Quote from: PoulGrym on July 20, 2011, 04:02:23 AM

What ever it was you did.. it's working.. Rejected shares have been very low! <1%

Just need to get scoring working.Can't have those hoppers jet as soon it's not worth it anymore.. We need to stick to it.

Agreed on the scoring. I have left for mineco.in until someone confirms the scoring is working as advertised. Hopefully that will be soon.

Quote from: kr105 on July 20, 2011, 04:15:16 AM

Awesome pool, i'll switch from deepbit as soon as the current round (on your pool) finish!

If the scoring is, in fact, working, then there isn't really any benefit to waiting until the end of a round to join the pool.

kr105

hero member

Activity: 938

Merit: 501

Awesome pool, i'll switch from deepbit as soon as the current round (on your pool) finish!

PoulGrym

member

Activity: 83

Merit: 10

What ever it was you did.. it's working.. Rejected shares have been very low! <1%

Just need to get scoring working.Can't have those hoppers jet as soon it's not worth it anymore.. We need to stick to it.

Meni Rosenfeld

donator

Activity: 2058

Merit: 1054

Quote from: Inaba on July 19, 2011, 01:14:50 PM

How does giveaways increase the variance? I wish I had a better math/stats background.

Depends on the kind of giveaway. For the case of 0.5 BTC reward to the block solver, it means that 99% of your expected payout is according to the scoring method (which has, say, 0.1% of solo variance), and 1% of it is effectively solo with 100% of solo variance. So your total variance is 1.1% of solo variance which is a X11 increase.

Inaba

legendary

Activity: 1260

Merit: 1000

Ok, I will have to give that some consideration. How does giveaways increase the variance? I wish I had a better math/stats background.

Meni Rosenfeld

donator

Activity: 2058

Merit: 1054

Quote from: Inaba on July 19, 2011, 12:55:16 PM

Just trying to bring the fee down to zero without destabilizing the variance too much.

Then just use f=-c/(1-c). All I said is that it might be problematic for you to have high variance and no average fee to show for it. If you're up to it then go for it. The giveaway ideas just cause the variance to be higher with no benefit.

FWIW, having zero fee is one of the last things I care about in a pool. Things like stability, website features, and low variance are much more important, and I'd be happy to pay 1%-2% fee if I know it incentivizes the operator to make the pool as good as possible. Just look at Tycho, he made the largest pool (hence lowest variance) by, for example, taking fees and using them for Google ads (not that I support this particular practice...).

Inaba

legendary

Activity: 1260

Merit: 1000

Just trying to bring the fee down to zero without destabilizing the variance too much.

Meni Rosenfeld

donator

Activity: 2058

Merit: 1054

Quote from: Inaba on July 19, 2011, 12:30:35 PM

Maybe a 5 BTC jackpot randomly selecting from the top share producers or perhaps block finders for the past 10 blocks?

What are you trying to solve again?

Inaba

legendary

Activity: 1260

Merit: 1000

Maybe a 5 BTC jackpot randomly selecting from the top share producers or perhaps block finders for the past 10 blocks?

Sorry about the website being down - I made a change to an include file and neglected to test it before walking away to a meeting. Didn't affect shares or miners, just the web front end. One day I will learn to test the website before walking away.

Meni Rosenfeld

donator

Activity: 2058

Merit: 1054

Quote from: rearwheels on July 19, 2011, 11:47:20 AM

Just put the 1% as bonus for the block finder.

No, this just makes the variance higher. The parameters to play with are f and c.

rearwheels

sr. member

Activity: 444

Merit: 254

Miners are still ok.

Only webfront is down.

Topic: [1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration! - page 208. (Read 499702 times)