[DEAD] DeepBit.net PPS+Prop,instant payouts, we pay for INVALID BLOCKS too - page 82.

DutchBrat

hero member

Activity: 868

Merit: 1000

Quote from: rjk on April 18, 2012, 07:28:20 AM

Quote from: kano on April 18, 2012, 07:11:03 AM

Quote from: DutchBrat on April 18, 2012, 06:42:18 AM

Isn't the race to be the first to get the LP also a lottery like who get's to deliver the solution to the hash ?

i.e. If you have 100 rigs setup with 1 GPU you have 100 lottery tickets to be the first to get a LP as opposed to 20 rigs with 5 GPU's where you have only 20 tickets....

For the really short rounds (in the hundreds/few thousands of shares) your hashing power isn't so much what counts as it is the time in which you get the LP (and then of course deliver a solution before the round ends)

Hmm what was the comment I made in IRC 9 hours ago to a pool OP ...

Quote

13:05 < kanoi> so ... ***** ... how does the pool decide the order it sends out LPs ...
(yes that is a rather controversial question, but only if the answer isn't truly random

I know that Slush prioritizes the LPs he sends out based on hashrate - I don't know whether Deepbit does this or not, but I think Slush is one of the only ones that does that.

So that effectively means that the higher your hash rate the more chance you have of submitting a share in a very short round.... so smaller miners will have less of a chance to participate in a 'more profitable' round than big miners. That means quick rounds don't even out based on variance... so the model becomes statistically biased as big miners will have a bigger chance to begin with (of course you have to factor in geographics / network speed / etc etc)

Edit: Mind you, it is all mostly a hypothetical discussion as there aren't that many rounds where not all miners participate (on Deepbit) and even though it would be nice to get a 'huge' payoff in a round from time to time, I have the feeling that not participating in such a round won't affect your overall 'expected' payout that much....

P4man

hero member

Activity: 518

Merit: 500

Bitminter does it too. Then again, with 100-150 GH there arent that many 1 minute rounds.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: rjk on April 18, 2012, 07:28:20 AM

I know that Slush prioritizes the LPs he sends out based on hashrate - I don't know whether Deepbit does this or not, but I think Slush is one of the only ones that does that.

Interesting. Prioritising by hashrate is probably a good way to reduce overall stales.

rjk

sr. member

Activity: 448

Merit: 250

1ngldh

Quote from: kano on April 18, 2012, 07:11:03 AM

Quote from: DutchBrat on April 18, 2012, 06:42:18 AM

Isn't the race to be the first to get the LP also a lottery like who get's to deliver the solution to the hash ?

i.e. If you have 100 rigs setup with 1 GPU you have 100 lottery tickets to be the first to get a LP as opposed to 20 rigs with 5 GPU's where you have only 20 tickets....

For the really short rounds (in the hundreds/few thousands of shares) your hashing power isn't so much what counts as it is the time in which you get the LP (and then of course deliver a solution before the round ends)

Hmm what was the comment I made in IRC 9 hours ago to a pool OP ...

Quote

13:05 < kanoi> so ... ***** ... how does the pool decide the order it sends out LPs ...
(yes that is a rather controversial question, but only if the answer isn't truly random

I know that Slush prioritizes the LPs he sends out based on hashrate - I don't know whether Deepbit does this or not, but I think Slush is one of the only ones that does that.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: kano on April 18, 2012, 07:08:31 AM

Quote from: organofcorti on April 18, 2012, 06:38:58 AM

Why is this? Workers on different accounts wont receive LPs sequentially afaik.

Unless I've completely missed something about non-broadcast TCP/IP packets travelling around the internet ...
... how can all the network packets arrive at the same time?

Even if the pool has multiple threads sending out the packets, they still have a sequential ordering through at least some the devices as they travel across the net - and of course the pool itself will not have as many threads as there are miners, that will be sending out the LP notices.

Having more accounts doesn't mean you're going to increase the average time to get LPs, which is what I thought you were saying in the post. It just means a reduction in overall LP variance.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: DutchBrat on April 18, 2012, 06:42:18 AM

Isn't the race to be the first to get the LP also a lottery like who get's to deliver the solution to the hash ?

i.e. If you have 100 rigs setup with 1 GPU you have 100 lottery tickets to be the first to get a LP as opposed to 20 rigs with 5 GPU's where you have only 20 tickets....

For the really short rounds (in the hundreds/few thousands of shares) your hashing power isn't so much what counts as it is the time in which you get the LP (and then of course deliver a solution before the round ends)

That's pretty much the way I see it along with delays caused by geographical distance. I did some analysis here and found that I could model the LP response fairly well (although I wasn't sure it was due to LP at the time, I am now) . The response time was distributed as a log normal distribution, mean 0.7 seconds, sd 1.65 seconds.

So I think I see Kano's point now - LP variations will have an effect on your variance for very short rounds. If the model above holds, the mean lag time is about 7.5 seconds, median 2.01 seconds and a standard deviation of about 30. It's quite skewed, half of the miners getting LPs before 2 seconds, and the rest receiving the LPs from 2 seconds up to around 100 seconds after the start of the round.

Edit: This isn't taking into account the time for the share to get from the miner to the pool. Also, the mode is 0.13 seconds, so most miners will have received their LPs by then.

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

Quote from: DutchBrat on April 18, 2012, 06:42:18 AM

Isn't the race to be the first to get the LP also a lottery like who get's to deliver the solution to the hash ?

i.e. If you have 100 rigs setup with 1 GPU you have 100 lottery tickets to be the first to get a LP as opposed to 20 rigs with 5 GPU's where you have only 20 tickets....

For the really short rounds (in the hundreds/few thousands of shares) your hashing power isn't so much what counts as it is the time in which you get the LP (and then of course deliver a solution before the round ends)

Hmm what was the comment I made in IRC 9 hours ago to a pool OP ...

Quote

13:05 < kanoi> so ... ***** ... how does the pool decide the order it sends out LPs ...
(yes that is a rather controversial question, but only if the answer isn't truly random

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

Quote from: organofcorti on April 18, 2012, 06:38:58 AM

Quote from: kano on April 18, 2012, 06:31:59 AM

Quote from: organofcorti on April 18, 2012, 05:51:54 AM

So how does that change if you have several accounts?

Well since the discussion started about getting shares in short rounds ...
and both you and DeepBit have stated that the number of accounts doesn't matter ...
My comments were pointing out that the hash rate is not the ONLY factor in determining if you will get shares in the short rounds.

Quote

A miner can't do much about LPs. Just making sure BlackPrapor realises this.

However, now that you mention it, if you have more accounts, then you will have to receive more LP messages.
This will slow down the first item I listed above - such that the average time to get the LP for each account will increase since every account after the first one to receive the LP will of course receive it later.

Why is this? Workers on different accounts wont receive LPs sequentially afaik.

Unless I've completely missed something about non-broadcast TCP/IP packets travelling around the internet ...
... how can all the network packets arrive at the same time?

Even if the pool has multiple threads sending out the packets, they still have a sequential ordering through at least some the devices as they travel across the net - and of course the pool itself will not have as many threads as there are miners, that will be sending out the LP notices.

DutchBrat

hero member

Activity: 868

Merit: 1000

Isn't the race to be the first to get the LP also a lottery like who get's to deliver the solution to the hash ?

i.e. If you have 100 rigs setup with 1 GPU you have 100 lottery tickets to be the first to get a LP as opposed to 20 rigs with 5 GPU's where you have only 20 tickets....

For the really short rounds (in the hundreds/few thousands of shares) your hashing power isn't so much what counts as it is the time in which you get the LP (and then of course deliver a solution before the round ends)

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: kano on April 18, 2012, 06:31:59 AM

Quote from: organofcorti on April 18, 2012, 05:51:54 AM

So how does that change if you have several accounts?

Well since the discussion started about getting shares in short rounds ...
and both you and DeepBit have stated that the number of accounts doesn't matter ...
My comments were pointing out that the hash rate is not the ONLY factor in determining if you will get shares in the short rounds.

Quote

A miner can't do much about LPs. Just making sure BlackPrapor realises this.

However, now that you mention it, if you have more accounts, then you will have to receive more LP messages.
This will slow down the first item I listed above - such that the average time to get the LP for each account will increase since every account after the first one to receive the LP will of course receive it later.

Why is this? Workers on different accounts wont receive LPs sequentially afaik.

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

Quote from: organofcorti on April 18, 2012, 05:51:54 AM

So how does that change if you have several accounts?

Well since the discussion started about getting shares in short rounds ...
and both you and DeepBit have stated that the number of accounts doesn't matter ...
My comments were pointing out that the hash rate is not the ONLY factor in determining if you will get shares in the short rounds.

However, now that you mention it, if you have more accounts, then you will have to receive more LP messages.
This will slow down the first item I listed above - such that the average time to get the LP for each account will increase since every account after the first one to receive the LP will of course receive it later.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

So how does that change if you have several accounts?

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

And yet ... that is ignoring a few of the facts that do affect this:
Firstly there is a delay from when the pool gets a block until it informs each miner of that new block via an LP - and the 'miner' receives the LP
Secondly there is the delay from that LP to the time when the first share is expected to be found based on the mining power
Thirdly there is the delay in sending the share back to the pool

The miner will have an LP delay + expected time + reply delay that is either within the short block time or not ...

... and a VERY specific example would be a BFL mining device that will not return a share for LP delay + 5.x seconds + reply delay ...

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: jkminkov on April 18, 2012, 02:09:16 AM

Quote from: organofcorti on April 16, 2012, 10:22:27 AM

BlackPrapor you requested someone work out the variance you can expect for you? I'll go one better.

If you have a 6970, I'm guessing you have a hashrate of 350 Mhps? Then you're submitting on average 4.88 difficulty 1 shares per minute. At the current pool hashrate of 3500Ghps, 50000 shares will take 61 seconds. Since the number of shares you submit in a given time is a poisson distributed variable, both the mean and the variance of your submissions in 61 seconds = 4.97. Let's round that to 5.

That might not be helpful, but below is a table of probabilities that you will submit a given number of shares in 61 seconds:

Code:

Shares Probability
0 0.006737947
1 0.033689735
2 0.084224337
3 0.140373896
4 0.175467370
5 0.175467370
6 0.146222808
7 0.104444863
8 0.065278039
9 0.036265577
10 0.018132789

At current difficulty the probability of a round lasting 25000 to 75000 shares is about 0.03. At the current pool hashrate, a block will be solved on average every 32 minutes, or 315 rounds per week. So in a week you'd see 10 rounds between 25000 and 75000 shares.

So you see it's not unlikely, but certainly not usual, that in a week there would be a 50000 share round in which you'd submit no shares.

take a note, that some miners use 2 threads per GPU(cgminer by default), so your numbers need recalculation I think :]

No, for this analysis the way shares are created isn't important - only the total you send to the pool in a given period of time is. If you send more shares than your effective hashrate is higher.

jkminkov

hero member

Activity: 698

Merit: 500

Quote from: organofcorti on April 16, 2012, 10:22:27 AM

BlackPrapor you requested someone work out the variance you can expect for you? I'll go one better.

If you have a 6970, I'm guessing you have a hashrate of 350 Mhps? Then you're submitting on average 4.88 difficulty 1 shares per minute. At the current pool hashrate of 3500Ghps, 50000 shares will take 61 seconds. Since the number of shares you submit in a given time is a poisson distributed variable, both the mean and the variance of your submissions in 61 seconds = 4.97. Let's round that to 5.

That might not be helpful, but below is a table of probabilities that you will submit a given number of shares in 61 seconds:

Code:

Shares Probability
0 0.006737947
1 0.033689735
2 0.084224337
3 0.140373896
4 0.175467370
5 0.175467370
6 0.146222808
7 0.104444863
8 0.065278039
9 0.036265577
10 0.018132789

At current difficulty the probability of a round lasting 25000 to 75000 shares is about 0.03. At the current pool hashrate, a block will be solved on average every 32 minutes, or 315 rounds per week. So in a week you'd see 10 rounds between 25000 and 75000 shares.

So you see it's not unlikely, but certainly not usual, that in a week there would be a 50000 share round in which you'd submit no shares.

take a note, that some miners use 2 threads per GPU(cgminer by default), so your numbers need recalculation I think :]

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: BlackPrapor on April 18, 2012, 12:20:18 AM

Quote from: DeepBit on April 17, 2012, 04:11:16 AM

Number of accounts won't make any difference. All shares are equal anyway.

You're answering like a real politician Grin

OK then: Your variance depends on total shares submitted. It doesn't matter if you submit them from one account or many.

Quote from: BlackPrapor on April 18, 2012, 12:20:18 AM

Quote from: DeepBit on April 17, 2012, 04:11:16 AM

Measuring hashrate with your last 7 shares won't give any anything even remotely correct.
Set your averaging window to 30-40 minutes and you'll see better results.

I imagine my stated hash rate for each of the last 7 blocks should be somewhere around my real hash rate. I can setup averaging window to see my reported speed for the last X nimutes, but its still deviates alot in the last 7 blocks column. I point to this fact because I get paid based on those numbers. I guess it's just another variable which doesn't affect your income in the long run.

Your round hashrate (in Mhps) is:

Code:

shares submitted for round * 2^32/10^6/time for round(in seconds)

Remember that the probability mass function (probabilities I gave above) is quite 'shallow' so variance is large. If the pool hashrate is stable, then your variance in round hashrate will be the same as your variance in shares submitted in a given round. For larger hashrates, the Poisson distribution variance is comparatively lower than for a low hashrate. So you're going to see significant variances in hashrate if you're looking at what the pool reports.

Asking about the variance of your shares submitted in a minute is the same as asking about the variance of your hashrate in a round.

BlackPrapor

hero member

Activity: 628

Merit: 504

Quote from: DeepBit on April 17, 2012, 04:11:16 AM

Number of accounts won't make any difference. All shares are equal anyway.

You're answering like a real politician Grin

Quote from: DeepBit on April 17, 2012, 04:11:16 AM

Measuring hashrate with your last 7 shares won't give any anything even remotely correct.
Set your averaging window to 30-40 minutes and you'll see better results.

I imagine my stated hash rate for each of the last 7 blocks should be somewhere around my real hash rate. I can setup averaging window to see my reported speed for the last X nimutes, but its still deviates alot in the last 7 blocks column. I point to this fact because I get paid based on those numbers. I guess it's just another variable which doesn't affect your income in the long run.

DeepBit

donator

Activity: 532

Merit: 501

We have cookies

Quote from: macbook-air on April 17, 2012, 10:13:49 AM

Hello, I am getting "Account_locked" while trying to create a new worker or change name of an existing worker. Why? I can still make manually payout.

Probably it was accidentally locked because of poolhopping. PM me your login name.

macbook-air

sr. member

Activity: 324

Merit: 260

Hello, I am getting "Account_locked" while trying to create a new worker or change name of an existing worker. Why? I can still make manually payout.

DeepBit

donator

Activity: 532

Merit: 501

We have cookies

Quote from: BlackPrapor on April 17, 2012, 12:03:29 AM

All I'm asking is whether I'll be able to reduce variance by having many accounts, and getting a share in a small block at least with some of those accounts.

Number of accounts won't make any difference. All shares are equal anyway.

Quote from: BlackPrapor on April 17, 2012, 12:03:29 AM

P.S. BTW, I'm using proportional payouts and my hash speed for the last 7 shares reported on account page is quite weird. I get anything from 190Mhs up to 933Mhs per round. My card's speed is stable though, at 362,8Mhs (not overclocked)

Measuring hashrate with your last 7 shares won't give any anything even remotely correct.
Set your averaging window to 30-40 minutes and you'll see better results.

Topic: [DEAD] DeepBit.net PPS+Prop,instant payouts, we pay for INVALID BLOCKS too - page 82. (Read 1601330 times)