[1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration! - page 54.

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Quote from: Inaba on September 28, 2012, 02:47:29 PM

Ok, here's the config now:

US1: 12 getworks per minute target
US2: 16 getworks per minute target
US3: 20 getworks per minute target

See which one works best for you.

So, mining with 1 5970 and 2 BFL mini rigs the results are:

US1 @ 12 getworks == diff between 50 and 75
US2 @ 16 getworks == diff between 40 and 50
US3 @ 20 getworks == diff between 30 and 40

I am really liking US3 with almost 1/2 the diff of US1. This has definitely helped bring share variance down and the swings in the diff changes as the server re-targets the diff.

Inaba

legendary

Activity: 1260

Merit: 1000

Ok, here's the config now:

US1: 12 getworks per minute target
US2: 16 getworks per minute target
US3: 20 getworks per minute target

See which one works best for you.

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Quote from: Inaba on September 28, 2012, 12:28:54 PM

Sure, I've chosen the values arbitrarily at this point, because I wasn't sure what optimal values would be.

Why do you choose 20, just out of curiosity? I chose 8 at first, but that seemed to really crank the difficulty, then 12... we can move to 20 and see how that plays out.

20 would allow for larger miners to have less variance in share submission which is usually the reward for have a larger hash rate at Diff 1. For 1Gh/s, share submission should be around 13.85 shares per minute. By having a number higher than 13.85, it still allows larger miners to have slightly less share variance while still limiting the overall pool resources the large miners are taking up.

P_Shep

legendary

Activity: 1795

Merit: 1208

This is not OK.

Quote from: Inaba on September 28, 2012, 12:28:54 PM

Sure, I've chosen the values arbitrarily at this point, because I wasn't sure what optimal values would be.

Why do you choose 20, just out of curiosity? I chose 8 at first, but that seemed to really crank the difficulty, then 12... we can move to 20 and see how that plays out.

You should find the difficulty which lowers the traffic to a minimum, whilst also minimising stales and maximizing hashrate (obvioulsy).
So pick a value to start with, then vary that (up and down) to minimise GW/m + shares/m.
The whole point of this Vardiff, is afterall, to minimise traffic.

Have a play, build a table, see what results you get.

Inaba

legendary

Activity: 1260

Merit: 1000

Sure, I've chosen the values arbitrarily at this point, because I wasn't sure what optimal values would be.

Why do you choose 20, just out of curiosity? I chose 8 at first, but that seemed to really crank the difficulty, then 12... we can move to 20 and see how that plays out.

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Is there a particular reason the var diff server tries to keep shares submission to 12 per minute? With diff 1 shares, as your hash rate increases your variance goes down but this is not the case with 12 shares per minute rule. Can this be set to 20? A share submission target of once every 3 seconds?

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

Quote from: jamesg on September 27, 2012, 02:27:52 PM

Quote from: -ck on September 27, 2012, 01:52:47 PM

I haven't decided what to do with the efficiency metric. Either I'll make up something or just not use it.

Is the efficiency metric still important with variable diff shares? I'm sitting at 62% currently.

Yes but as I said on the previous page - the variable diff version of that can be calculated in the API
However ...

You can also find the two numbers on the screen with: P I n Enter (where n is the pool number)
So from there divide "Accepted difficulty shares:" / "Getworks"

The00Dustin

hero member

Activity: 807

Merit: 500

Quote from: Roy Badami on September 27, 2012, 03:49:31 PM

I think this is how it works:

In order to submit the block solo, you'd have to change the address to which the block reward is payable to be one that you control. And you'd have to do this before solving the block, otherwise you'd change the hash. A pool operator obviously isn't going to accept a share where you've modified the coinbase transaction to change the address the block reward is paid to - so your shares would simply be rejected.

I had to think about it for a few seconds, but that makes sense, even if you can change the address in the template, once you do, you are solo mining, and if you don't, the block can't pay you. Even simpler.

Roy Badami

hero member

Activity: 563

Merit: 500

Quote from: The00Dustin on September 27, 2012, 01:46:12 PM

Quote from: Inaba on September 27, 2012, 01:39:26 PM

You have the template for the pools block, not a solo block... so it has to be submitted through the pool to be a valid block, otherwise the key won't match the template.

Makes sense, so to be much more specific, what the pool has that the miner doesn't is the pool's private key for the address being generated against, correct?

I think this is how it works:

In order to submit the block solo, you'd have to change the address to which the block reward is payable to be one that you control. And you'd have to do this before solving the block, otherwise you'd change the hash. A pool operator obviously isn't going to accept a share where you've modified the coinbase transaction to change the address the block reward is paid to - so your shares would simply be rejected.

Yes, it's true that the one thing the pool operator knows but you don't is the private key of the address that the block reward is payable to - but that's really answering a different question, namely: "why can't you spend someone else's coins?"

roy

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Quote from: -ck on September 27, 2012, 01:52:47 PM

I haven't decided what to do with the efficiency metric. Either I'll make up something or just not use it.

Is the efficiency metric still important with variable diff shares? I'm sitting at 62% currently.

-ck

legendary

Activity: 4088

Merit: 1631

Ruu \o/

Quote from: Inaba on September 27, 2012, 12:28:21 PM

Don't forget about GBT, either. It basically reduces the outbound traffic (replacing Getwork) to 1 connection every 2 minutes regardless of your hashrate.

That said, how does that effect efficiency calculations going forward? Stratum is effectively the same in that regard, so if you pull a template and send back getworks, how is CGminer going to calculate efficiency or does that just become a redundant metric at that point?

Quote from: -ck on September 27, 2012, 03:20:04 AM

Anyway soon with stratum and friends all these statistics become irrelevant.

That's what I was referring to.

I haven't decided what to do with the efficiency metric. Either I'll make up something or just not use it.

The00Dustin

hero member

Activity: 807

Merit: 500

Quote from: Inaba on September 27, 2012, 01:39:26 PM

You have the template for the pools block, not a solo block... so it has to be submitted through the pool to be a valid block, otherwise the key won't match the template.

Makes sense, so to be much more specific, what the pool has that the miner doesn't is the pool's private key for the address being generated against, correct?

Inaba

legendary

Activity: 1260

Merit: 1000

You have the template for the pools block, not a solo block... so it has to be submitted through the pool to be a valid block, otherwise the key won't match the template.

The00Dustin

hero member

Activity: 807

Merit: 500

Quote from: Inaba on September 27, 2012, 12:51:18 PM

You are still submitting shares (at the appropriate variable difficulty) - you just don't need to request work constantly as you already have the block template and you build the block on your end, instead of the pool building it for you. Every two minutes or a LP will get you an updated template to build the block off of.

Just out of curiosity, I have wondered for a long time, but given this basic description of GBT, it is time to ask. Is there something the pool has (which would be hard to regenerate) that the miner doesn't (beyond the block template)? I should imagine there is, because surely I'm not the only one who would think that otherwise a block witholding attack could becomea block stealing attack if the miner can submit the real blocks on their own since they have the template.

Inaba

legendary

Activity: 1260

Merit: 1000

You are still submitting shares (at the appropriate variable difficulty) - you just don't need to request work constantly as you already have the block template and you build the block on your end, instead of the pool building it for you. Every two minutes or a LP will get you an updated template to build the block off of.

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Quote from: Inaba on September 27, 2012, 12:28:21 PM

Don't forget about GBT, either. It basically reduces the outbound traffic (replacing Getwork) to 1 connection every 2 minutes regardless of your hashrate.

That said, how does that effect efficiency calculations going forward? Stratum is effectively the same in that regard, so if you pull a template and send back getworks, how is CGminer going to calculate efficiency or does that just become a redundant metric at that point?

This is the one piece I don't fully understand yet. How does GBT let the pool know how fast I am hashing so the pool knows what proportion of blocks to give me? This is pretty straight forward with getwork as I am submitting shares whether they are winners or not.

Inaba

legendary

Activity: 1260

Merit: 1000

Don't forget about GBT, either. It basically reduces the outbound traffic (replacing Getwork) to 1 connection every 2 minutes regardless of your hashrate.

That said, how does that effect efficiency calculations going forward? Stratum is effectively the same in that regard, so if you pull a template and send back getworks, how is CGminer going to calculate efficiency or does that just become a redundant metric at that point?

jamesg

vip

Activity: 1358

Merit: 1000

AKA: gigavps

Just moved ~50Gh in eclipse and the variable diff stuff is awesome! The network definitely has a lot less traffic now.

P_Shep

legendary

Activity: 1795

Merit: 1208

This is not OK.

I haven't coded up Anubis to display that yet (get to it soon), but the stats say that maxbtc has roll time of 60, and oz/emc of 120. So I was expecting similar GW to oz.

kano

legendary

Activity: 4592

Merit: 1851

Linux since 1997 RedHat 4

Quote from: P_Shep on September 27, 2012, 03:00:31 AM

Peaks at 9... There's no way EMC some how had over 20000 shares (equivilent) submitted in the same time the others has 10000.
Forget about submitted shares.
I'm just talking about Get works. EMC has more get works over the same time than the other pools. That shouldn't be the case.

Again you need to look at the API to get conclusive numbers to compare pool performance (or any performance) due to the effect of difficulty.
Yes the number of GetWorks is relevant, but the figures I mentioned before MUST be used in the comparison to ensure that EMC is being compared properly (since EMC can have >1 difficulty shares also)
Also, the API stats report has information in it that may need to be considered in a comparison also (e.g. roll-n-time info)
I'm not saying any numbers are right or wrong, but simply that those API report figures must be looked at in any comparison.

Topic: [1200 TH] EMC: 0 Fee DGM. Anonymous PPS. US & EU servers. No Registration! - page 54. (Read 499597 times)