[1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration! - page 48.

kano

legendary

Activity: 4634

Merit: 1851

Linux since 1997 RedHat 4

Quote from: organofcorti on October 10, 2012, 10:33:06 AM

...
Unless you plan on making the relationship completely linear and reducing pool D to less than one? It would be a great selling point for the tinier miners.

Although cgminer already works with LTC mining and thus handles fractional D, it wont actually work in any BTC miner (or cgminer) since they all check for H=0 thus you'd lose valid shares if D < 1

Seriously, that's a hard limit in all BTC mining since all BTC devices only return 1 Diff shares (H=0)

That's a big part of the 6.25% performance optimisation in GPU mining ... and anyone who reads this, and is thinking of suggesting all the mining devices and GPU OCL code be rewritten, simply doesn't understand the negative implications or the time wasted in trying to do that

Askit2

hero member

Activity: 981

Merit: 500

DIV - Your "Virtual Life" Secured and Decentralize

Hey Inaba could you check the email system. The last email I had delivered to inbox or junk mail was on Sunday.
Thank You!

The00Dustin

hero member

Activity: 807

Merit: 500

Quote from: organofcorti on October 10, 2012, 09:50:56 AM

Quote from: The00Dustin on October 10, 2012, 09:43:42 AM

I am assuming letting miners pick their target is a losing option based solely on my opinion of human nature.

Seems to work ok at HHTT.

I assume ASICs don't exist yet. I have seen a lot of posts about "the biggest miners donate the least." IF the biggest miners are that selfish, it seems like a safe assumption that they will choose lower diff in order to decrease variance for the same reasons they donate the least. I know this is an assumption, hence the comment: "based solely on my opinion of human nature."

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: Inaba on October 10, 2012, 10:08:15 AM

It looks like 20 - 24 is an optimal range to satisfy the 1 minute emotional need for the vast majority of people... am I reading that right?

I think whatever you use the 1 minute variance is going to be something people will worry about. At 20 shares per minute the 95% CI for average per minute hashrate is -40% to +45%, and at 24 shares per minute the 95% CI for average per minute hashrate is -38% to +42%. At 1Ghps and difficulty 1 you're only submitting an average of 13.98 shares per minute anyway.

I'll try to find a way to compare variance miners already experience at pool D = 1 to what they'd experience at another D or number of shares per minute.

Otherwise I think the variance in share submission rates a 1Ghps miner experiences at pool D = 1 should be fine for most miners. Say 14 shares per minute to make it easier to calculate pool D. It also makes it easier for miners to know when the variable pool D begins - at 1Ghps and below pool D = 1, and above 1Ghps pool D starts to increase.

Unless you plan on making the relationship completely linear and reducing pool D to less than one? It would be a great selling point for the tinier miners.

Inaba

legendary

Activity: 1260

Merit: 1000

It looks like 20 - 24 is an optimal range to satisfy the 1 minute emotional need for the vast majority of people... am I reading that right?

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: The00Dustin on October 10, 2012, 09:43:42 AM

So a better conclusion (and I'm still in this thread because I am trying to understand your work only as it applies to his pool right now) would be "if one wants to use a target number of shares per minute, he can only refer to the graphs in the first section to cherry-pick a variation range."

Yes. Or the table. Or the CSV files (can't post links here unfortunately)

Quote from: The00Dustin on October 10, 2012, 09:43:42 AM

I am assuming letting miners pick their target is a losing option based solely on my opinion of human nature.

Seems to work ok at HHTT.

The00Dustin

hero member

Activity: 807

Merit: 500

So a better conclusion (and I'm still in this thread because I am trying to understand your work only as it applies to his pool right now) would be "if one wants to use a target number of shares per minute, he can only refer to the graphs in the first section to cherry-pick a variation range." I am assuming letting miners pick their target is a losing option based solely on my opinion of human nature.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: Luke-Jr on October 10, 2012, 09:04:39 AM

Quote from: The00Dustin on October 10, 2012, 08:58:05 AM

Would "a target of 13 shares per minute would keep variance most consistent with a 1gh/s connection" be a valid conclusion? I gathered this primarily from the last chart in your blog. I don't know if that is Inaba's goal anyway, but someone mentioned a desire for something to that effect.

The exact number is 13.969838619232177734375 shares per minute: 60 seconds ÷ (2³² nonces-per-pdiff1-share-on-average ÷ 1,000,000,000 hashes-per-second)

OK then:

A target of 13.969838619232177734375 shares per minute would keep variation in average average per minute hashrate to -50% to 57.5%, variation in average hourly hashrate to ~ +/- 6.7% and variation in average daily hashrate to +/- 1.4%, all with 95% confidence.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: The00Dustin on October 10, 2012, 08:58:05 AM

Would "a target of 13 shares per minute would keep variance most consistent with a 1gh/s connection" be a valid conclusion? I gathered this primarily from the last chart in your blog. I don't know if that is Inaba's goal anyway, but someone mentioned a desire for something to that effect.

This probably belongs over in the Neighbourhood Pool Watch thread, but anyway your answer is no it would not be a valid conclusion. If Inaba plans on having a set difficulty per minute for all miners then the variation in average hashrate does not depend on the miner hashrate.

This case is covered in the first table in section two.

A better conclusion would be "a target of 13 shares per minute would keep variation in average average per minute hashrate to -55% to 64%, variation in average hourly hashrate to ~ +/- 7% and variation in average daily hashrate to +/- 1.4%, all with 95% confidence.

HTH.

Luke-Jr

legendary

Activity: 2576

Merit: 1186

Quote from: The00Dustin on October 10, 2012, 08:58:05 AM

Would "a target of 13 shares per minute would keep variance most consistent with a 1gh/s connection" be a valid conclusion? I gathered this primarily from the last chart in your blog. I don't know if that is Inaba's goal anyway, but someone mentioned a desire for something to that effect.

The exact number is 13.969838619232177734375 shares per minute: 60 seconds ÷ (2³² nonces-per-pdiff1-share-on-average ÷ 1,000,000,000 hashes-per-second)

The00Dustin

hero member

Activity: 807

Merit: 500

Would "a target of 13 shares per minute would keep variance most consistent with a 1gh/s connection" be a valid conclusion? I gathered this primarily from the last chart in your blog. I don't know if that is Inaba's goal anyway, but someone mentioned a desire for something to that effect.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

I had a bit more spare time than I thought I would, so I finished the post a little early. It's a bit of a rush job so please let me know if there are any words missing, typos, spelling errors etc. Post is here: http://organofcorti.blogspot.com/2012/10/71-variable-pool-difficulty.html

If you have any comments or questions about it, please post a comment on the blog or leave a post here: https://bitcointalksearch.org/topic/neighbourhood-pool-watch-66026

FLHippy

full member

Activity: 784

Merit: 101

Quote from: mdude77 on October 09, 2012, 07:20:07 PM

Quote from: FLHippy on October 09, 2012, 07:07:14 PM

Quote from: mdude77 on October 09, 2012, 06:54:42 PM

Interesting. How long did you do this test?

M

About an hour in each configuration.

I don't think an hour is a fair test. Maybe a week would shake out the inherent sine curve that tends to haunt mining.

M

OOf... I keep expecting next week to be the end of gpu mining :/ I'm running on borrowed time and every day brings less BTC

I think I'm just going to leave it where it is and just forget about it.

FLHippy

full member

Activity: 784

Merit: 101

Quote from: kano on October 09, 2012, 09:21:00 PM

Yes that was sarcasm.

mean people suck.

kano

legendary

Activity: 4634

Merit: 1851

Linux since 1997 RedHat 4

Quote from: FLHippy on October 09, 2012, 07:07:14 PM

Quote from: mdude77 on October 09, 2012, 06:54:42 PM

Interesting. How long did you do this test?

M

About an hour in each configuration.

Hmm - I'll have to go find that hour where one of my rigs ran 30% above expected results and count that as proof that the pool was better than every other pool in existence ...

Yes that was sarcasm.

Seriously, 1 hour means almost nothing in terms of working out which is best for you ...

The simple fact is that with higher difficulty you will get higher variance.
Standard BTC 101

None of the 3 servers will get you more or less expected BTC in the long term.

Edit: what you are looking for is something along the lines of:
If my hash rate is X GH/s what difficulty would give me an expected variance similar to hashing at 1GH/s with 1 difficulty shares.
... and that's a maths problem - not an observation problem.

organofcorti

donator

Activity: 2058

Merit: 1007

Poor impulse control.

Quote from: kano on October 09, 2012, 03:13:41 PM

Didn't organofcorti post somewhere the appropriate calculation?
(for vardiff)

It's somewhere in the p2Pool thread.

I'm flat out like a lizard drinking atm, but I'll do an NPW post in the next few days with charts and look up tables based on either a miner selecting a pool difficulty, or a pool selecting a number of shares per minute for all miners.

mdude77

legendary

Activity: 1540

Merit: 1001

Quote from: FLHippy on October 09, 2012, 07:07:14 PM

Quote from: mdude77 on October 09, 2012, 06:54:42 PM

Interesting. How long did you do this test?

M

About an hour in each configuration.

I don't think an hour is a fair test. Maybe a week would shake out the inherent sine curve that tends to haunt mining.

M

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Quote from: FLHippy on October 09, 2012, 06:30:35 PM

I did some testing this afternoon using three workers on PPS.

Numbers:

combined daily earnings reported by website for all devices on 1 worker on us3. 0.61 BTC/day
combined daily earnings reported by website for one worker per device using us1,us2,us3 0.64 BTC/day
combined daily earnings reported by website for one worker per device using us3. 0.69 BTC/day

Hash rates reported by web server were 1.90Gh/s, 1.98GH/s, and 2.10GH/s

This is because us1 has the highest "share" variance because the server is targeting you to submit less shares per minute by using a higher diff. The more shares you submit per minute, the lower your variance.

FLHippy

full member

Activity: 784

Merit: 101

Quote from: mdude77 on October 09, 2012, 06:54:42 PM

Interesting. How long did you do this test?

M

About an hour in each configuration.

mdude77

legendary

Activity: 1540

Merit: 1001

Quote from: FLHippy on October 09, 2012, 06:30:35 PM

I did some testing this afternoon using three workers on PPS.

I have 650-700MH/s per card. All cards are MSI R 7970 cards. I'm using BFGMiner.

I set up a worker for each card and pointed them at us1, us2, and us3.

To get it to run on different servers and to isolate each card I ran a separate instance of bfgminer for each card (Even the 2 which are on the same server)

with my rig, us3 produces the best return and I believe us1 produced the lowest return.

Also, all three devices on one worker produced lower returns than three workers on three servers.

Numbers:

combined daily earnings reported by website for all devices on 1 worker on us3. 0.61 BTC/day
combined daily earnings reported by website for one worker per device using us1,us2,us3 0.64 BTC/day
combined daily earnings reported by website for one worker per device using us3. 0.69 BTC/day

Hash rates reported by web server were 1.90Gh/s, 1.98GH/s, and 2.10GH/s

Interesting. How long did you do this test?

M

Topic: [1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration! - page 48. (Read 499791 times)