Quote from: baddw on August 29, 2014, 07:10:25 PM

Quote from: luigi1111 on August 29, 2014, 05:24:19 PM

Quote from: baddw on August 29, 2014, 04:00:18 PM

Quote from: luigi1111 on August 29, 2014, 02:42:24 PM

Quote from: baddw on August 29, 2014, 05:41:13 AM

Quote from: luigi1111 on August 29, 2014, 02:28:55 AM

Quote from: baddw on August 29, 2014, 12:10:53 AM

Quote from: thelonecrouton on August 28, 2014, 11:39:19 PM

Sporkage is just putting the rules you mention into action. Same as if they were hardcoded in the first place.

Of course they had to be hardcoded in the first place for the spork to even work, but the fact that the software can be influenced by an external "switch" that can be turned on or off at any time is not in the spirit of cryptocurrencies. I am fine with it for betas/RCs/development/whatever, but the final product should definitely not have any spork switches.

Quote from: luigi1111 on August 28, 2014, 11:46:29 PM

I don't believe enforcement was ever difficult to implement, I think the problems resulted from the payment/election code sometimes causing different nodes to believe different masternodes should be paid. So when enforcement was switched on (via hard fork), little forks started emerging because there wasn't consensus.

That is what I understand to have been happening; could've been something else though.

This is certainly believable, and I guess it would be a problem if a given miners/pools "chose" the same MNs over and over again, excluding other MNs. 1. However, it is better for them to be paying MNs than not at all.

Personally I've always thought the "election" process was convoluted and prone to error (given that not all MNs can necessarily always be seen by all nodes on the network) and that the coinbase transaction should simply pay the MNs who provably performed the mixing during that particular block. Split it up among them proportionately. The clients will choose which MNs to send to at random; or if they mess with the code to always favor certain MNs, that's fine too (although they will probably be losing some privacy if they choose to always use their own MN = shooting themselves in the foot = hopefully nobody is so stupid as to do this). The MNs are paid for providing a service, and if they provide that service in a given block, then they should be paid in that block. JM2C

1. I believe this should have been a fairly easy "enforcement" switch to code, but in reality wouldn't have done too much. I believe the cause of the forking to have been nodes thinking different MNs should be paid, and rejecting the blocks that paid, but didn't pay the "right" node. However, just making so 20% had to be split off or the block rejected would allow the miner to just pay themselves in a separate transaction.

I think you have an interesting idea in doing that PoSvc thing (which I believe "flare" has talked about at dct): you should be able to determine which masternode signed each denomination transaction included in the block.

It sounds very cool and elegant at first, but it too has vectors that must be considered. Nothing stops you from choosing a masternode; they just sit around waiting for inputs. So you could constantly denominate the smallest # of inputs (to minimize fees) using your own node, which would nearly guarantee its inclusion in every block for payment. This only wouldn't make sense if fees were higher than the payout.

Additionally, what happens if no denominate transaction occurs in a particular block? Is the mining node then able to keep the entire reward for himself? What then happens if he simply refused to include those transactions in his block (I'm not sure how you can force them to include transactions?).

Several other ideas are rolling around in my head, mostly in relation to masternodes somehow/sometimes having block minting authority. I need to think about it some more though.

Ok, good points.... keeping up a synchronized, agreed-upon list of MNs throughout the network is the real crux here. Especially since it definitely needs to account for MNs dropping out of the network from time to time. Which shouldn't really happen, but it will happen for various reasons. So yeah, that idea definitely has some flaws that would be hard to work through.

How about this. Each block, a group of MNs is "nominated" by the block hash. The hash simply matches a few bits of the MN's public address... target this so that each block hash should "nominate" 5-10 MNs on average. Then all of the nominated MNs will be required to create "ping" transactions and push them to the network within 2 blocks. These transactions will be included in these next 2 blocks. Take the hash of the 2nd block after the original "nomination" block, and use it to determine the winner (among the nominees who have correctly "pinged" back) of the MN prize that will be generated by the coinbase transaction in the next, 3rd block. If for some reason a block hash doesn't match any MNs (or no MNs happen to "ping" in time -- which is equivalent in result), then there would be a fallback option such as awarding the prize to a different MN that wasn't chosen in the previous block, but was nominated and "pinged" back correctly.

This gets rid of the problem of having to keep a current, up-to-date list of all MNs throughout the network. The MNs themselves are responsible for letting the network know that they are out there by the "ping" transactions, and they can be called upon at any time due to the random nature of the nomination process, so therefore they must be ready to quickly respond at any time. If they miss a "ping" then they will be penalized by not being eligible for the reward the next time they are nominated. All MN-payment-eligible addresses will have their public addresses in the blockchain immediately prior (1 or 2 blocks) to the block which pays them out, and the reward will be awarded to one of the nominees in a deterministic fashion, which eliminates confusion and rejection of blocks due to nodes not being aware of a particular MN on the network.

There might be some problems that crop up with this process due to orphan blocks. Perhaps the ping-response-time should be set to 5 or even 10 blocks instead of 2. But I have been chewing on this kind of idea for a while now, and I think it could work for the MNs.

Well as fate would have it, I didn't dream about this at all...

It seems to me that the distribution of masternode public keys is probably not very uniform, and therefore a (pseudo) random matching based on block hashes could result in rather uneven payments? It may be possible to mitigate this, though not 100% sure. Perhaps some additional variables in the matching algorithm could help. If it can be structured in a way to get a reasonably normal distribution, then it's actually rather elegant.
One other consideration is that it'd still be highly probable that producing an output not matching any MNs is possible. Do we just say "too bad" for those ineligible blocks?

Match on several bits along the length of the entire public address. The address is a hash function. Its output should be sufficiently random.

I addressed the "no match" problem above. If there isn't a match, choose a winner among the non-winning addresses from the previous block. If there was only one address in that block, then go back to the previous block, recursively until a block is found with multiple addresses, and choose one of the non-winning addresses from that block, using the same choosing algorithm.

I am thinking that this choosing algorithm might simply be an ordering. Like say the block hash used to choose the winning address among the nominee addresses is A7Bg34xyz. Take the first digit in this hash, the 7. This means choose the 7th character in the nominee addresses. Sort the nominee addresses by this 7th character (and carry over to the 8th character to settle ties). Now that they are ordered, take the second digit in the hash, the 3. This means choose the 3rd nominee address as the winner. If there aren't 3 qualified nominees, choose the 2nd, or the 1st, etc.

The algorithm for nominating MNs could be similar, to provide for added randomness. Take the same block hash, A7Bg34xyz. Take the first digit in the hash, 7, and the next character after it, B. This tells us that the 7th character in the address needs to be a 'B'. Now go to the next digit in the hash, 3, and take the next character after it, 4. This tells us that the 3rd character in the address needs to be a '4'. So all MNs whose public addresses match the pattern xx4xxxBxxxxxxxx would be nominated by that block. This algorithm could obviously be tweaked in many ways in order to generate the right number of MN addresses with each block. Maybe the pattern match would need to be broadened to be any pattern of 4xxxB anywhere in the address. Or only in the first 15 characters of the address. Or the last 15 characters of the address. Etc.

There are a lot of variations to play with. But it is hard for me to see a way for anybody to game this somehow by generating their address in order to be paid more or less often. I guess if they had multiple MNs then they could generate the address of each so that there is no overlap with their other addresses, so they would never be nominated at the same time as the others. Which would be ok. If there was a lot of overlap then they could have multiple MNs nominated at once, which would increase their chances of winning the payment for a block in which they have multiple nominated MNs. So it's hard for me to see a real winner there.... it seems to be equal. Better chances of winning when nominated, versus being nominated more often. Given that the process for both nomination and choosing the winner are essentially random as long as no single entity is controlling the mining, I don't see how you could argue that one of these options is definitely better than the other.

I like it. I'm curious how much junk it'd add to blocks with "alive pings", etc. If no MNs matched/pinged from the previous block(s), you could potentially pay a MN that had been offline for a while, but I think that's a relatively minor issue (you couldn't really code it out either, as requiring "re-pinging" would take an extra block, leaving you with no one to pay for the present block).

You still need the old MN list though, for accounting and verification purposes (hot/cold would also be impossible without it). The block finding node needs to verify all the pings are from acceptable masternodes before including them in his block. This seems like it could take us back to old sync problem, BUT since the list wouldn't need to keep track of whether or not the MNs were alive, only that they were capable of existing (i.e., signed message from DRK address with 1,000 input tx designating new MN pubkey), perhaps the sync issue would be mitigated?

That leaves us with the possible conundrum of verification discrepancies. The other nodes have to verify his block against the list or the block finder could include addresses that aren't MNs (his own made up addresses; it shouldn't be hard at all to generate on-the-fly addresses that fit the requirements with 5/1,000 expected to fit). Does the paying block finder (n+2 in your proposal I think) also verify his payee is in the list? He wouldn't need to if all the peers had verified the n block as being valid, but that could potentially bring us the forking problem again. I think he should, particularly if it's a "reused block", but his peers don't need to/shouldn't; more below (sorry this is getting kind of scattered).

I think there are compromises that could make it more than acceptable. Consider: what if there are "allowances" in the "ping" block to cover potential discrepancies in the list? Like 75% or so of the included pings must exist in the list or peers would reject it. Something like that should make potential forks from list discrepancies very rare at worst, non-existent at best. Then, at block n+2, the payer could again verify his payee is in the list (so he doesn't pay the "naughty" address the n finder included to try to pay himself), but his peers wouldn't (obviously payee must be from block n), avoiding forks over list discrepancies there! Additionally, in the case of a "rerun", the payer would be able to check his payee is still a valid node.

For this to work, the payee wouldn't be deterministic, but chosen at random from n by n+2 block finder. He could of course pay himself, but only if he was in the list to start with (so it seems very marginally beneficial to him). The other consideration is that if "Pool A" found both n and n+2, he could get away with paying whoever, but that seems minor to me again (others might disagree).

Thoughts? Perhaps an MN list without the "alive" state included would be stable enough so as to not cause many (if any) discrepancies and thus forks, and all my thinking is for naut? (It's fun following these trains of thought regardless though Grin

)

Edit: sorry for WoT folks; baddw and I are just out on the playground here.

No, you don't really need the list anymore. The "ping" transaction must be signed by a valid MN holding 1000 DRK. Verifying this would be no more time-consuming than verifying any other transaction. (If I try to spend 100BTC from an address that doesn't have 100BTC, this transaction is rejected by any and every node on the network. If I try to make a MN "ping" from an address with less than 1000 DRK and/or that has not previously announced itself as a cold-wallet MN, it will be rejected as well. If a bad miner tries to include a "ping" from an address that is not a verified valid MN, this block will be rejected for containing an invalid transaction.)

The paying block finder only has to verify that his payee has sent a valid "ping" transaction within the past 2 blocks. Again, the "list" doesn't come into play. If you wanted to create a "list" of MNs then you would simply scan the blockchain for all valid "ping" transactions within the past X amount of time (2-3 days, maybe?) And you would drop any who should have responded to a more recent "nomination" but didn't. After a full scan of the blockchain, you would have a full list of all MNs ever seen on the network, and when is the last time that they should have pinged, and when is the last time that they actually pinged.

Trying to keep a complete, correct, fully-up-to-date "list" in real-time is an exercise in futility. Without some centralized gatekeeper, the updated list cannot be propagated and agreed across the network quickly enough to reliably keep up with MN payments and not have some of them rejected by some nodes. It is this difficulty in immediate and constant consensus that the blockchain was created to solve. The blockchain can provide a method to create a reliable-enough, complete-enough, up-to-date-enough "list" of un-rejectable MNs much as it is already used to create a "list" of un-rejectable unspent outputs.

Am I missing something here? Where is such an "announcement" stored? A masternode can only sign its ping with its provided private key, which has no relation to a DRK address holding 1k except via signed message by the actual address giving authority for said masternode key to act on its behalf.

Perhaps the solution is to have additional space in the blockchain for storing these messages (this is what you were thinking all along, isn't it (you're a genius!))? They'd exist for all eternity unless pruned at some point in the future (though that would need some workaround to keep the blocks paying them in the past valid when the evidence of their being a masternode is removed), but I'm not sure that matters... You would actually be able to make a "list" of sorts then based on the messages in the blockchain, but it wouldn't vary at all from node to node! Assume the same address could initiate a stop command in a future block to not have his old MN clutter up the payments by never responding. However, I don't see any incentive for people to do this, so there needs to be some mechanism where the block finder can insert the stop message on behalf of the masternode. You might think the lingering nodes wouldn't be an issue, but something absolutely has to be able to issue a "stop" command in case the 1,000 DRK move. Unless you could possibly modify the protocol to have nodes reject any tx formed out of the 1,000 DRK input without a "masternode stop" command included in a prior block? If you allow for removal after X number of missed pings, you have the (tiny) vector of the block finder being able to remove a masternode if he found the right blocks, but I don't know of any incentive to do this.

Now we're just stuck with n block finder (that happens to have his node in the list of candidates) pretending he didn't get pings from some or all of the other candidates. Perhaps this vector is small enough to just discard.

This is actually sounding like a workable solution! Everyone give it up for baddw! (*crickets*, no one is listening to us banter)

I wrote most of this before fully comprehending what you wrote in your last paragraph, which is basically suggesting the same thing I'm going over, right?

Edit: by "list", I only mean abstractly in the form the blockchain provides/would provide..

Edit2: some typos, also sorry again thread for the huge mess this has become Tongue

Edit3: just noticed in a previous post (quoted above) I spelled "naught" as "naut". Subliminal coin advertising, anyone? Grin