[2.5+ EH] Slush Pool (slushpool.com); World's First Mining Pool

[Boognish]

Can anyone explain the benefit of DGM over PPLNS?  Pool hopping is also prevented by PPLNS and seems like it would be much easier to implement.  From what I gather, DGM helps 'absorb' a little bit of variation but I would imagine that isn't as big a deal on a larger pool like Slush's. 

[1529406402]

1529406402

[1529406402]

1529406402

[1529406402]

1529406402

[1529406402]

1529406402

[organofcorti]

Can anyone explain the benefit of DGM over PPLNS?  Pool hopping is also prevented by PPLNS and seems like it would be much easier to implement.  From what I gather, DGM helps 'absorb' a little bit of variation but I would imagine that isn't as big a deal on a larger pool like Slush's. 

PPLNS is a subset of DGM solutions. DGM has much more flexibility than PPLNS: variance and maturity time can be tuned to the miners' and pool op's needs and wants. PPLNS cannot.

All of the info (not just my interpretation of it) is here.

Btw there are a number of DGM pools now, so it can't be too much harder than PPLNS to implement here - especially when Slush has more experience than anyone at running a pool using exponential scoring).

[slush]

...so it can't be too much harder than PPLNS to implement here - especially when Slush has more experience than anyone at running a pool using exponential scoring).

The main difference between current score implementation is that score equation needs only time as an input, but proper implementation of DGM need to know also count of already submitted shares (across all backends). Pool core is designed quite scalable, there's no online synchronization between backends necessary, but it goes against DGM requirements. And it's the pain and the reason why it takes so much time. Originally I expect that it won't be so hard, but I felt into solving strange threading issues (cause framework which I'm currently using doesn't natively support user's threads).

[organofcorti]

The main difference between current score implementation is that score equation needs only time as an input, but proper implementation of DGM need to know also count of already submitted shares (across all backends). Pool core is designed quite scalable, there's no online synchronization between backends necessary, but it goes against DGM requirements. And it's the pain and the reason why it takes so much time. Originally I expect that it won't be so hard, but I felt into solving strange threading issues (cause framework which I'm currently using doesn't natively support user's threads).

Thanks for your input there - it's interesting to know how the differences in payout implementation affect pool operators.

Would moving to PPLNS temporarily be a possibility? Or would you be changing one difficult to solve problem for another?

[slush]

Would moving to PPLNS temporarily be a possibility? Or would you be changing one difficult to solve problem for another?

At this moment is simplier to finish DGM than do all coding and testing for PPLNS and then DGM again...

[slush]

I'll start linux kernel & database update in following 30 minutes. Unfortunately I need to stop application, expected downtime will be 20-30 minutes.

[slush]

Ok, it's back up, downtime was 35 minutes.

[jothan]

...so it can't be too much harder than PPLNS to implement here - especially when Slush has more experience than anyone at running a pool using exponential scoring).

The main difference between current score implementation is that score equation needs only time as an input, but proper implementation of DGM need to know also count of already submitted shares (across all backends). Pool core is designed quite scalable, there's no online synchronization between backends necessary, but it goes against DGM requirements. And it's the pain and the reason why it takes so much time. Originally I expect that it won't be so hard, but I felt into solving strange threading issues (cause framework which I'm currently using doesn't natively support user's threads).

Can't you multiply (or add the logs) of all the "s" variables from all the backend nodes and then do the same for the per-user scores (sk) ?

It might work, but I suck at math, Meni may have an idea on how to algebrize that.

Edit: Just in case I was not clear, the idea would be to have each backend node do DGM independently and then multiply all the backend "s" together and all the backend "sk" together to estimate or compute the payout.

[mem]

pool appears to be offline atm

still first time Ive ever seen an outage on slush, finally had to configure a failover pool in cgminer.

[slush]

Writing from mobile connection; i was fixing one issue on the server, sorry for short outage. Connection from here was incredibly laggy. I also stopped payouts, they'll be back on the evening UTC.

[splitt]

Hi. I noticed that recently there have been a lot of invalid blocks here on Slush.
Is that normal or just bad luck?
What is the average number of invalid blocks for, say, 30 days?

Thanks!

[slush]

splitt, invalid ratio is slightly under 1% in long term average. Yes, in recent days there were some extra invalids, but as you can trace on blockchain.info, it was really just bad luck, there isn't any obvious mistake. Pool is also directly connected to some well propagated nodes and also to few other pools to minimize latency.

[slush]

Payouts are still disabled, I came at home too late and it's always better to be fresh while handling with coins. I'm working on fix right now and payouts should be enabled very soon.

[slush]

OK, BTC payouts working again (I just sent batch of all pending payouts), I'm working on NMC now.

Edit: NMC payouts works as well.

[Meni Rosenfeld]

Can anyone explain the benefit of DGM over PPLNS?  Pool hopping is also prevented by PPLNS and seems like it would be much easier to implement.  From what I gather, DGM helps 'absorb' a little bit of variation but I would imagine that isn't as big a deal on a larger pool like Slush's. 

PPLNS can work too, but there are a few differences to consider:
1. What most people refer to as PPLNS is the naive variant which isn't completely hopping-proof. There are correct variants but they're slightly harder to implement.
2. PPLNS is potentially more resource-intensive, as it requires a history of shares - DGM only strictly requires keeping track of the score of every miner.
3. DGM can absorb variance, which may or may not be needed or used but it's nice to have the option.
4. In DGM the decay is gradual rather than abrupt, this can be seen as a pro or a con depending on taste.

There's also shift-PPLNS and shift-DGM which work better with a parallel architecture.

[jothan]

There's also shift-PPLNS and shift-DGM which work better with a parallel architecture.

Could you explain how shift-PPLNS and shift-DGM work ? I have seen you use that expression before, but I have never seen you explain it. Are you refering to the version of DGM where the logarithms of the scores are used ?  Using 2 as a logarithm base, addition of exponents would be equivalent to binary shifting.

Can you have distributed DGM with per-node worker scores and "s" variable and then add all the nodes together later ?

[Meni Rosenfeld]

I have seen you use that expression before, but I have never seen you explain it.

That's because I was never asked

Shift-PPLNS is a method invented by DrHaribo of BitMinter which I then refined a bit. I've discussed it briefly in this comment as well as section "PPLNS variants" of AoBPMRS.

Shift here refers not to bit-shift, rather to worker shifts. Shares are grouped into shifts, and miners are rewarded for the shares they submitted in last few shifts rather than the last N shares. Since all shares submitted within a shift are equivalent, there's no need to store the individual shares, rather just the total shares submitted per miner per shift (or more accurately the total pB).

By using a gap of 1 shift between work and reward, it also doesn't matter (on average) whether a share is counted as belonging to one shift or the next. So this works in parallel - each pool instance tallies work per shift for the miners connected to it (because of the gap, it is ok to be a little out of sync with the other instances wrt which is the current shift) and then compute rewards after the fact with the summed data.

Shift-DGM is applying the same principles to DGM. I've never actually worked out the details (again, due to lack of interest), but it should work. The idea is that shares will decay only between shifts rather than within a shift, so you can do it in parallel with out-of-sync instances. Within a shift it's a simple tally of the total pB of shares submitted.

[slush]

My appologies for recent downtime, it was database maintenance (second and final part). Downtime was less than 10 minutes.

[crunchyferrett]

My appologies for recent downtime, it was database maintenance (second and final part). Downtime was less than 10 minutes.

Was EXACTLY 10 minutes by my watch. But who is counting?
:p

[ludo0777]

This is a great pool!