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

[Inaba]

Thanks! Should be fixed now!

[chrcoe01]

it's all working for me now yay!  probably has been for the past day or so, but I haven't really checked it until now :/

[-ck]

So Gigavps and anyone else:  Did we ever decide on what a good share target was?  20?  24?

I'd say 10 shares per minute would be fine.

[FLHippy]

So Gigavps and anyone else:  Did we ever decide on what a good share target was?  20?  24?

I'd say 10 shares per minute would be fine.

I agree... 10 shares per minute seems like a reasonable number of network connections.

I have a very general question about variable difficulty which has bugged me.

I've asked before and I think the answer was experiment and see how you make out. I did some of that and the earnings have steadily been dropping due to increased hash rates, bad luck, and increased difficulty. So... it is hard to tell the results of testing.

So the question is...

If you have multiple devices... should each device have its own worker or should all of your devices share a worker?

If the devices are FPGA/GPU - should you use a different approach than when people receive ASIC hardware?

[Inaba]

So Gigavps and anyone else:  Did we ever decide on what a good share target was?  20?  24?

I'd say 10 shares per minute would be fine.

I was at 8 and that was too low... I moved it up to 16 and that still seems too low for some people.  Of course, when the ASICs are out, I think 10 is absolutely reasonable, if the minimum hashrate on a given unit is 4.5 GH/s.  Right now, though, I think 10 might be too low for GPU miners... not from a technical perspective, but from an emotional one: it drives their variance up too high for comfort is the feeling I get from people.

I think we might need to go back to the drawing board and shoot for a variable difficulty based on server load, vs a getwork target... though that adds quite a bit of complexity.  I'm not sure what metric would be the best to account for server load, as there are many other factors that come into play just looking at the system load in top or some such.

I have a very general question about variable difficulty which has bugged me.

I've asked before and I think the answer was experiment and see how you make out. I did some of that and the earnings have steadily been dropping due to increased hash rates, bad luck, and increased difficulty. So... it is hard to tell the results of testing.

So the question is...

If you have multiple devices... should each device have its own worker or should all of your devices share a worker?

If the devices are FPGA/GPU - should you use a different approach than when people receive ASIC hardware?

Well, the FPGA/GPU vs ASIC question really needs to be asked as at what GH/s speed does the getwork target make the most sense... so if you combine all your units into one worker, then a lower getwork target makes more sense, since your variance will "apparently" be reduced by the higher hashrate.  If you split them all up, a higher target is better, for the same reason.  It's all about perception for the most part... over a long enough period, it doesn't really matter from a functional standpoint.

[ragnard]

I'm a GPU miner with about 600MH/s on one machine under one worker.  I haven't seen my Diff go higher than 1 during this whole testing phase, so it doesn't appear to be affecting my variance at all.

[jamesg]

I was at 8 and that was too low... I moved it up to 16 and that still seems too low for some people.  Of course, when the ASICs are out, I think 10 is absolutely reasonable, if the minimum hashrate on a given unit is 4.5 GH/s.  Right now, though, I think 10 might be too low for GPU miners... not from a technical perspective, but from an emotional one: it drives their variance up too high for comfort is the feeling I get from people.

I think we might need to go back to the drawing board and shoot for a variable difficulty based on server load, vs a getwork target... though that adds quite a bit of complexity.  I'm not sure what metric would be the best to account for server load, as there are many other factors that come into play just looking at the system load in top or some such.

I would really like to see a scenario where:

• Pool server checks for the X-Mining-Hashrate header and calculates a diff based on the reported hash rate similar to what conman originally suggested. The lowest diff is still 1.
• If X-Mining-Hashrate doesn't exist, fall back to 20-30 shares per minute per worker target.
• If the server load is too high, move the lowest diff allow from 1 to 2.

What do you think?

[P_Shep]

As I mentioned a few pages back, do experimentations, gather data, build a table of results, draw graphs, then chose a method that works best for you. Pulling numbers out of the air is not a scientific method. Get numbers.

[kano]

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

[Inaba]

I must have missed it if he did, anyone got a lead on it?

[FLHippy]

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

[mdude77]

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

[FLHippy]

Interesting.  How long did you do this test?

M

About an hour in each configuration.

[jamesg]

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.

[mdude77]

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

[organofcorti]

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.

[kano]

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.

[FLHippy]

Yes that was sarcasm.

mean people suck.

[FLHippy]

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.

[organofcorti]

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://bitcointalk.org/index.php?topic=66026.0