[1500 TH] p2pool: Decentralized, DoS-resistant, Hop-Proof pool

[stevegee58]

What kind of discarded work rates are people getting with cgminer on p2pool?

I read that it's normal on p2pool to have a higher discarded work rate than on a traditional pool.  I'm just trying to figure out how much is normal to see if I have problems or not.
My discarded work is about 1.5x the accepted.

[forrestv]

What kind of discarded work rates are people getting with cgminer on p2pool?

I read that it's normal on p2pool to have a higher discarded work rate than on a traditional pool.  I'm just trying to figure out how much is normal to see if I have problems or not.
My discarded work is about 1.5x the accepted.

That's definitely not normal. You should get at most 2% of shares rejected, as displayed in cgminer. If you're getting anything more, upgrade cgminer and decrease its intensity.

[stevegee58]

I wasn't referring to the rejected field but the "DW" field which is discarded work.  My reject rate is 0.9%.

The DW field increments every time cgminer shows a long poll but not an accept.

[DeathAndTaxes]

What kind of discarded work rates are people getting with cgminer on p2pool?

I read that it's normal on p2pool to have a higher discarded work rate than on a traditional pool.  I'm just trying to figure out how much is normal to see if I have problems or not.
My discarded work is about 1.5x the accepted.

That's definitely not normal. You should get at most 2% of shares rejected, as displayed in cgminer. If you're getting anything more, upgrade cgminer and decrease its intensity.

Discarded work isn't rejected.  It is simply work cgminer requested and then never starting working before discarded due to LP so it doesn't "cost" you anything.

Either because of settings or a bug cgminer is requesting a lot more work from p2pool then it will be able to finish.  When next LP occurs that requested work becomes stale so cgminer discards it (without doing any work on it).  It then requests a crap ton of work it will never be able to finish and 10 seconds later the same thing happens.

If you have threads per GPU > 1 and/or queue > 1 then change them both to 1.  Also knock the intensity down 1 or 2 notches.  

With a "normal" pool I find the optimal settings for a 5970 to be:
Queue: 2
Threads per GPU: 2
Intensity: 9

For p2pool I find the optimal settings for a 5970 to be:
Queue: 1
Threads per GPU: 1
Intensity: 8

LP are so short that maintaining a large queue of work is mostly wasted.  With an average time between LP of 10 seconds you need to be complete > 4.3 million hashes or you will never complete more than 1 getwork per LP.   That works out to ~430 MH/s (per GPU).  Using higher queue and/or threads really only makes sense if someone if using a > 430 MH/s miner.  Not per rig, or card but per chip.  Using both queue and threads >1 only makes sense on some hypothetical  > 1 GH/s chip.

[forrestv]

I wasn't referring to the rejected field but the "DW" field which is discarded work.  My reject rate is 0.9%.

The DW field increments every time cgminer shows a long poll but not an accept.

Ah, oops, thought you were talking about rejects.. Anyway, it isn't harmful. Mine is about 80%.

[stevegee58]

I'm running a 6770 at around 200 MH/s with 1 queue and 1 thread.  Also I use dynamic intensity which usually sits around 4-5.
Sounds like my settings are right where they should be.

I'm thinking this discarded work metric isn't really that useful anyway.

Discarded work isn't rejected.  It is simply work cgminer requested and then never starting working before discarded due to LP.  

Still it isn't "normal".  The good news is high discarded work shouldn't be costing you shares.  It is "work" that cgminer never started.  It is for some reason requesting more work from p2pool then it could ever possibly finish and then discarding most of that before ever starting.

If you have threads per GPU > 1 and/or queue > 1 then change them both to 1.  Also knock the intensity down 1 or 2 notches.  With a "normal" pool I find the optimal settings for a 5970 to be
Queue: 2
Threads per GPU: 2
Intensity: 9

For p2pool I find the optimal settings for a 5970 to be:
Queue: 1
Threads per GPU: 1
Intensity: 8


LP are so short that maintaining a large queue of work is mostly wasted.

To complete nonce range in <10 seconds requires ~430 MH/s (per GPU).  Using higher queue and/or threads really only makes sense if someone if using a > 430 MH/s miner.  Not per rig but per chip.  Using both queue and threads >1 only makes sense on some hypothetical  > 1 GH/s chip.

[spiccioli]

What kind of discarded work rates are people getting with cgminer on p2pool?

I read that it's normal on p2pool to have a higher discarded work rate than on a traditional pool.  I'm just trying to figure out how much is normal to see if I have problems or not.
My discarded work is about 1.5x the accepted.

That's definitely not normal. You should get at most 2% of shares rejected, as displayed in cgminer. If you're getting anything more, upgrade cgminer and decrease its intensity.

[...]

With an average time between LP of 10 seconds you need to be complete > 4.3 million hashes or you will never complete more than 1 getwork per LP.   That works out to ~430 MH/s (per GPU).  Using higher queue and/or threads really only makes sense if someone if using a > 430 MH/s miner.  Not per rig, or card but per chip.  Using both queue and threads >1 only makes sense on some hypothetical  > 1 GH/s chip.

DeathAndTaxes,

on p2pool it would be usefull to have an option that makes cgminer split nonce ranges between GPUs so that the full 4 billion nonces can be tested during the available 10 seconds.

spiccioli.

[rjk]

I'm running a 6770 at around 200 MH/s with 1 queue and 1 thread.  Also I use dynamic intensity which usually sits around 4-5.
Sounds like my settings are right where they should be.

I'm thinking this discarded work metric isn't really that useful anyway.

In the context of P2Pool, no it isn't particularly useful.

[DeathAndTaxes]

DeathAndTaxes,

on p2pool it would be usefull to have an option that makes cgminer split nonce ranges between GPUs so that the full 4 billion nonces can be tested during the available 10 seconds.

spiccioli.

There wouldn't be a lot of value to that.  The only reason for wanting to check the entire nonce range is to reduce load on the server but w/ p2pool you are the server so a couple extra getworks per minute isn't going to really create any load issues.  As GPU/FPGA continue to get faster the it becomes less of an issue anyways.  A 7970 for example will need to process 2 nonce ranges per LP on average.  So will BFL & Icarus FPGAs.

[gnar1ta$]

Just saw a payment that was too high and found this

Code:

2012-02-24 21:29:38.976578 GOT BLOCK FROM MINER! Passing to bitcoind! http://blockexplorer.com/block/000000000000094d9ff2c8bfad49aeb36f569f4f9e5f9750f27924323eec244b

That's my first block on p2pool.

[kano]

Just saw a payment that was too high and found this

Code:

2012-02-24 21:29:38.976578 GOT BLOCK FROM MINER! Passing to bitcoind! http://blockexplorer.com/block/000000000000094d9ff2c8bfad49aeb36f569f4f9e5f9750f27924323eec244b

That's my first block on p2pool.

Hmm I wonder what that is:
value :   0.00077170 BTC
scriptPubKey :   OP_IFDUP OP_IF OP_2SWAP OP_VERIFY OP_2OVER OP_DEPTH

also Your coinbase (7) : 062f503253482f (./P2SH/)

[Rubberduckie]

Is the flag to change queue in cgminer -Q 1?

edit: just worked it out

[Diapolo]

1. I setup the bitcoin.conf and use the Windows GUI version 0.52 of the Bitcoin Client and forward TCP ports 6667 and 8333 to the Client.
2. I setup p2pool and forward TCP Port 9333 to the client.
3. I setup CGMINER to use -o http://127.0.0.1:9332 -u User_from_bitcoin.conf -p Password_from_bitcoin.conf.

CGMINER works, there are many LPs and a higher reject rate, than with a normal pool, which is okay from what I know.

But the question is now, how will the payouts work? Is there any good read or reference for this?
I'm asking, because in the p2pool window I see:

Code:

2012-02-25 11:52:42.775000  Shares: 0 (0 orphan, 0 dead) Stale rate: ??? Efficie
ncy: ??? Current payout: 0.0000 BTC

and

Code:

2012-02-25 11:53:16.471000  Local: 479MH/s in last 10.0 minutes Local dead on ar
rival: ~6.0% (2-15%) Expected time to share: 1.4 hours

Will it take 1.4 hours to see an estimated payout amount and will that directly show up in my Bitcoin Client GUI?

Thanks,
Dia

[Krak]

Will it take 1.4 hours to see an estimated payout amount and will that directly show up in my Bitcoin Client GUI?

Your estimated payout will start showing up after you've mined for a while. The payouts will go to a wallet in the Bitcoin client unless you specify a different address with the -a flag.

[Ed]

Just saw a payment that was too high and found this

Code:

2012-02-24 21:29:38.976578 GOT BLOCK FROM MINER! Passing to bitcoind! http://blockexplorer.com/block/000000000000094d9ff2c8bfad49aeb36f569f4f9e5f9750f27924323eec244b

That's my first block on p2pool.

did you receive block finder bounty?

[wachtwoord]

Will it take 1.4 hours to see an estimated payout amount and will that directly show up in my Bitcoin Client GUI?

Thanks,
Dia

When the p2pool finds a bitcoin block everyone that contributed will be paid in proportion. see http://p2pool.info/ for the times we found bitcoin blocks, how long the current one takes and how much payout you can expect if we find a block this instant.

The 1.4 hours is a (very rough) estimate of how long it will take you to find a p2pool block (based on your average hash rate in the last 10 minutes I believe). P2ppool blocks are far easier than a bitcoin block (the total p2pool finds one every 10 seconds) and these are used to determine who to pay what when the total p2pool finds a bitcoin block. This happnes proportionally to the total amount of p2pools people submitted in the 24 hours prior to the discovery of the bitcoin block.

Hope this clears things up

[Frizz23]

did you receive block finder bounty?

block finder bounty?

[wachtwoord]

did you receive block finder bounty?

block finder bounty?

The miner  which finds an actual block (main blockchain block) will receive a bonus (5% if I recall correctly), as an extra incentive to report winning blocks

[gnar1ta$]

Just saw a payment that was too high and found this

Code:

2012-02-24 21:29:38.976578 GOT BLOCK FROM MINER! Passing to bitcoind! http://blockexplorer.com/block/000000000000094d9ff2c8bfad49aeb36f569f4f9e5f9750f27924323eec244b

That's my first block on p2pool.

Hmm I wonder what that is:
value :   0.00077170 BTC
scriptPubKey :   OP_IFDUP OP_IF OP_2SWAP OP_VERIFY OP_2OVER OP_DEPTH

also Your coinbase (7) : 062f503253482f (./P2SH/)

Does that translate to yay I found a block and got some extra BTC? 

[twmz]

block finder bounty?

The miner  which finds an actual block (main blockchain block) will receive a bonus (5% if I recall correctly), as an extra incentive to report winning blocks

It also affects people who have bitcoin with a crappy network connection, because if all the blocks you find end up being orphaned because of a poor bitcoin configuration, you will not get the block finder bonus you were due.

Keep in mind that it's not an "extra payment" (as "bonus" might imply) since is taken out of the available 50 BTC for the block.  In other words, every non-block finder gets a little less than they would normally have if the entire block was being split.  Because blocks are found (over time) proportionally to the hashrate of the miners, it ends up being a wash in the long term.  You make a little less in the blocks you don't find, but you make more for any blocks you find, and the two offset each other over time.

So it ends up being neutral for people who are doing everything correctly and it is a penalty for anyone who is not doing things correctly (either bad bitcoin setup or intentionally withholding blocks).