we can all relax now
21 messages, 6 Nov 2013 to 10 May 2014. Started by Kyle Jerviss.
Peter Todd (5), Kyle Jerviss (4), Frank Flores, Jeff Garzik, Christophe Biocca, Melvin Carvalho, Jouke Hofman, Gavin Andresen, +6 more
One of the things that really gets me going is when someone devises a model, tests it against itself, and then pretends that they've learned something about the real world.
Naturally, the Selfish Mining paper is exactly this sort of nonsense. Their model is one with no latency, and one where the attacker has total visibility across the network. An iterated FSM is not a suitable simulation of the bitcoin system. The bitcoin network does not have states, and to the extent that you can pretend that we do, you can't simulate transitions between them with static probabilities.
The authors understand this deep down inside, even though they didn't work out the implications. They handwave the issue by assuming a total sybil attack, and in true academic spirit, they don't realize that the condition necessary for the attack is far, far worse than the attack itself.
Greg said he'd like to run some simulations, and I'm thinking about it too. Unfortunately, he is busy all week, and I'm lazy (and also busy for most of tomorrow).
If neither of us get to it first, I'm willing to pitch in 1 BTC as a bounty for building a general bitcoin network simulator framework. The simulator should be able to account for latency between nodes, and ideally within a node. It needs to be able to simulate an attacker that owns varying fractions of the network, and make decisions based only on what the attacker actually knows. It needs to be able to simulate this "attack" and should be generic enough to be easily modified for other crazy schemes.
(Bounty offer is serious, but expires in one year [based on the earliest timestamp that my mail server puts on this email], and /may/ be subject to change if the price on any reputable exchange breaks 1000 USD per BTC in that period.)
Basically, the lack of a decent network simulator is what allowed this paper to get press. If the author had been able to see the importance of the stuff he was ignoring, we wouldn't be wasting so much time correcting him (and sadly the reporters that have no way to check his claims).
https://bitcointalk.org/index.php?topic=324413.msg3495663#msg3495663
The problem with academics is that they don't have to worry about the real world. They get paid to publish things, not to be helpful to society.
On Tue, Nov 5, 2013 at 11:33 PM, kjj <bitcoin-devel@jerviss•org> wrote:
Quoted text (59 lines)
> One of the things that really gets me going is when someone devises a > model, tests it against itself, and then pretends that they've learned > something about the real world. > > Naturally, the Selfish Mining paper is exactly this sort of nonsense. > Their model is one with no latency, and one where the attacker has total > visibility across the network. An iterated FSM is not a suitable > simulation of the bitcoin system. The bitcoin network does not have > states, and to the extent that you can pretend that we do, you can't > simulate transitions between them with static probabilities. > > The authors understand this deep down inside, even though they didn't > work out the implications. They handwave the issue by assuming a total > sybil attack, and in true academic spirit, they don't realize that the > condition necessary for the attack is far, far worse than the attack > itself. > > Greg said he'd like to run some simulations, and I'm thinking about it > too. Unfortunately, he is busy all week, and I'm lazy (and also busy > for most of tomorrow). > > If neither of us get to it first, I'm willing to pitch in 1 BTC as a > bounty for building a general bitcoin network simulator framework. The > simulator should be able to account for latency between nodes, and > ideally within a node. It needs to be able to simulate an attacker that > owns varying fractions of the network, and make decisions based only on > what the attacker actually knows. It needs to be able to simulate this > "attack" and should be generic enough to be easily modified for other > crazy schemes. > > (Bounty offer is serious, but expires in one year [based on the earliest > timestamp that my mail server puts on this email], and /may/ be subject > to change if the price on any reputable exchange breaks 1000 USD per BTC > in that period.) > > Basically, the lack of a decent network simulator is what allowed this > paper to get press. If the author had been able to see the importance > of the stuff he was ignoring, we wouldn't be wasting so much time > correcting him (and sadly the reporters that have no way to check his > claims). > > https://bitcointalk.org/index.php?topic=324413.msg3495663#msg3495663 > > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development >
I will contribute 1 BTC to this bounty, under same terms and expiration.
I might try building this sometime soon. I think it may also serve an educational purpose when trying to understand the whole network's behaviour.
What level of accuracy are we looking for though? Obviously we need to fully emulate the steps of the network protocol, and we need to be able to specify time taken for transmission/processing for each node. Do we care about the actual contents of the messages (to be able to simulate double spend attempts, invalid transactions and blocks, SPV node communication), and their validation (actual signatures and proof of work)?
I imagine the latter is pretty useless, beyond specifying that the signature/proof of work is valid/invalid.
If we could build up a set of experiments we'd like to run on it, it would help clarify what's needed.
Off the top of my head:
- Peter Todd's miner strategy of sending blocks to only 51% of the hashpower. - Various network split conditions, and how aware of the split nodes would be (and the effect of client variability). - Testing the feasability of network race double spends, or Finney attacks. - Various network partition scenarios. - Tricking SPV nodes. On Nov 6, 2013 6:37 AM, "Jeff Garzik" <jgarzik@bitpay•com> wrote:
Quoted text (17 lines)
> I will contribute 1 BTC to this bounty, under same terms and expiration. > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development > >
On 6 November 2013 06:33, kjj <bitcoin-devel@jerviss•org> wrote:
Quoted text (43 lines)
> One of the things that really gets me going is when someone devises a > model, tests it against itself, and then pretends that they've learned > something about the real world. > > Naturally, the Selfish Mining paper is exactly this sort of nonsense. > Their model is one with no latency, and one where the attacker has total > visibility across the network. An iterated FSM is not a suitable > simulation of the bitcoin system. The bitcoin network does not have > states, and to the extent that you can pretend that we do, you can't > simulate transitions between them with static probabilities. > > The authors understand this deep down inside, even though they didn't > work out the implications. They handwave the issue by assuming a total > sybil attack, and in true academic spirit, they don't realize that the > condition necessary for the attack is far, far worse than the attack > itself. > > Greg said he'd like to run some simulations, and I'm thinking about it > too. Unfortunately, he is busy all week, and I'm lazy (and also busy > for most of tomorrow). > > If neither of us get to it first, I'm willing to pitch in 1 BTC as a > bounty for building a general bitcoin network simulator framework. The > simulator should be able to account for latency between nodes, and > ideally within a node. It needs to be able to simulate an attacker that > owns varying fractions of the network, and make decisions based only on > what the attacker actually knows. It needs to be able to simulate this > "attack" and should be generic enough to be easily modified for other > crazy schemes. > > (Bounty offer is serious, but expires in one year [based on the earliest > timestamp that my mail server puts on this email], and /may/ be subject > to change if the price on any reputable exchange breaks 1000 USD per BTC > in that period.) > > Basically, the lack of a decent network simulator is what allowed this > paper to get press. If the author had been able to see the importance > of the stuff he was ignoring, we wouldn't be wasting so much time > correcting him (and sadly the reporters that have no way to check his > claims). > > https://bitcointalk.org/index.php?topic=324413.msg3495663#msg3495663 >
Thanks for posting this bounty. I'm interested in working on it, and will give it a try. I also have some other commitments, so I suspect you guys will finish it first tho... but if not, I'll post details of the simulator.
Quoted text (17 lines)
> > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development >
bounty++
On 06-11-13 06:33, kjj wrote:
Quoted text (55 lines)
> One of the things that really gets me going is when someone devises a > model, tests it against itself, and then pretends that they've learned > something about the real world. > > Naturally, the Selfish Mining paper is exactly this sort of nonsense. > Their model is one with no latency, and one where the attacker has total > visibility across the network. An iterated FSM is not a suitable > simulation of the bitcoin system. The bitcoin network does not have > states, and to the extent that you can pretend that we do, you can't > simulate transitions between them with static probabilities. > > The authors understand this deep down inside, even though they didn't > work out the implications. They handwave the issue by assuming a total > sybil attack, and in true academic spirit, they don't realize that the > condition necessary for the attack is far, far worse than the attack itself. > > Greg said he'd like to run some simulations, and I'm thinking about it > too. Unfortunately, he is busy all week, and I'm lazy (and also busy > for most of tomorrow). > > If neither of us get to it first, I'm willing to pitch in 1 BTC as a > bounty for building a general bitcoin network simulator framework. The > simulator should be able to account for latency between nodes, and > ideally within a node. It needs to be able to simulate an attacker that > owns varying fractions of the network, and make decisions based only on > what the attacker actually knows. It needs to be able to simulate this > "attack" and should be generic enough to be easily modified for other > crazy schemes. > > (Bounty offer is serious, but expires in one year [based on the earliest > timestamp that my mail server puts on this email], and /may/ be subject > to change if the price on any reputable exchange breaks 1000 USD per BTC > in that period.) > > Basically, the lack of a decent network simulator is what allowed this > paper to get press. If the author had been able to see the importance > of the stuff he was ignoring, we wouldn't be wasting so much time > correcting him (and sadly the reporters that have no way to check his > claims). > > https://bitcointalk.org/index.php?topic=324413.msg3495663#msg3495663 > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development >
On Wed, Nov 06, 2013 at 01:06:47PM -0500, Christophe Biocca wrote:
Quoted text (20 lines)
> I might try building this sometime soon. I think it may also serve an > educational purpose when trying to understand the whole network's behaviour. > > What level of accuracy are we looking for though? Obviously we need to > fully emulate the steps of the network protocol, and we need to be able to > specify time taken for transmission/processing for each node. Do we care > about the actual contents of the messages (to be able to simulate double > spend attempts, invalid transactions and blocks, SPV node communication), > and their validation (actual signatures and proof of work)? > > I imagine the latter is pretty useless, beyond specifying that the > signature/proof of work is valid/invalid. > > If we could build up a set of experiments we'd like to run on it, it would > help clarify what's needed. > > Off the top of my head: > > - Peter Todd's miner strategy of sending blocks to only 51% of the > hashpower.
Speaking of, I hadn't gotten around to doing up the math behind that strategy properly; turns out 51% I was overly optimistic and the actual threshold is 29.3%
Suppose I find a block. I have Q hashing power, and the rest of the network 1-Q. Should I tell the rest of the network, or withhold that block and hope I find a second one?
Now in a purely inflation subsidy environment, where I don't care about the other miners success, of course I should publish. However, if my goals are to find *more* blocks than the other miners for whatever reason, maybe because transaction fees matter or I'm trying to get nLockTime'd announce/commit fee sacrifices, it gets more complicated.
There are three possible outcomes:
1) I find the next block, probability Q 2) They find the next block, probability 1-Q 2.1) I find the next block, probability Q, or (1-Q)*Q in total. 2.2) They find the next block, probability (1-Q)^2 in total.
Note how only in the last option do I lose. So how much hashing power do I need before it is just as likely that the other miners will find two blocks before I find either one block, or two blocks? Easy enough:
Q + (1-Q)*Q = (1-Q)^2 -> Q^2 - Q + 1/2 -> Q = (1 - \sqrt(2))/2
Q ~= 29.2%
So basically, if I'm trying to beat other miners, once I have >29.3% of the hashing power I have no incentive to publish the blocks I mine!
But hang on, does it matter if I'm the one who actually has that hashing power? What if I just make sure that only >29.3% of the hashing power has that block? If my goal is to make sure that someone does useless work, and/or they are working on a lower height block than me, then no, I don't care, which means my original "send blocks to >51% of the hashing power" analysis was actually wrong, and the strategy is even more crazy: "send blocks to >29.3% of the hashing power" (!)
Lets suppose I know that I'm two blocks ahead:
1) I find the next block: Q (3:0) 2) They find the next block: (1-Q) (2:1) 2.1) I find the next block: (1-Q)*Q (3:1) 2.2) They find the next block: (1-Q)^2 (2:2) 2.2.1) I find the next block: (1-Q)^2 * Q (3:2) 2.2.2) They find the next block: (1-Q)^3 (2:3)
At what hashing power should I release my blocks? So remember, I win this round on outcomes 1, 2.1, 2.2.1 and they only win on 2.2.2:
Q + (1-Q)*Q + (1-Q)^2*Q = (1-Q)^3 -> Q = 1 - 2^-3
Q ~= 20.6%
Interesting... so as I get further ahead, or to be exact the group of miners who have a given block gets further ahead, I need less hashing power for my incentives to be to *not* publish the block I just found. Conversely this means I should try to make my blocks propagate to less of the hashing power, by whatever means necessary.
Now remember, none of the above strategy requires me to have a special low-latency network or anything fancy. I don't even have to have a lot of hashing power - the strategy still works if I'm, say, a 5% pool. It just means I don't have the incentives people thought I did to propagate my blocks widely.
The other nasty thing about this, is suppose I'm a miner and recently got a block from another miner: should I forward that block, or not bother? Well, it depends: if I have no idea how much of the hashing power has that block, I should forward the block. But again, if my goal is to be most likely to get the next block, I should only forward in such a way that >30% of the hashing power has the block.
This means that if I have some information about what % already has that block, I have less incentive to forward! For instance, suppose that every major miner has been publishing their node addresses in their blocks - I'll have a pretty good idea of who probably has that most recent block, so I can easily make a well-optimized decision not to forward. Similarly because the 30% hashing power figure is the *integral* of time * hashes/second, if miners are forwarding near-target-headers, I might as well wait a few seconds and see if I see any near-target-headers; if I do for this block then I have evidence that hashing power does have it, and I shouldn't forward.
So yeah, we're fucked and have got to fix this awful incentive structure somehow before the inflation subsidy gets any smaller. Also, raising the blocksize, especially by just removing the limit, is utter madness given it can be used to slow down block propagation selectively, so the hashing power that gets a given block is limited repeatably to the same group.
P.S: If any large pools want to try this stuff out, give me a shout. You have my PGP key - confidentiality assured.
P.P.S: If you're mining on a pool with more than, like, 1% hashing power, do the math on varience... Seriously, stop it and go mine on a smaller pool, or better yet, p2pool.
You are ignoring the gambler's ruin. We do not operate on an infinite timeline. If you find a big pool willing to try this, please give me enough advance warning to get my popcorn ready.
Peter Todd wrote:
Quoted text (139 lines)
> On Wed, Nov 06, 2013 at 01:06:47PM -0500, Christophe Biocca wrote: >> I might try building this sometime soon. I think it may also serve an >> educational purpose when trying to understand the whole network's behaviour. >> >> What level of accuracy are we looking for though? Obviously we need to >> fully emulate the steps of the network protocol, and we need to be able to >> specify time taken for transmission/processing for each node. Do we care >> about the actual contents of the messages (to be able to simulate double >> spend attempts, invalid transactions and blocks, SPV node communication), >> and their validation (actual signatures and proof of work)? >> >> I imagine the latter is pretty useless, beyond specifying that the >> signature/proof of work is valid/invalid. >> >> If we could build up a set of experiments we'd like to run on it, it would >> help clarify what's needed. >> >> Off the top of my head: >> >> - Peter Todd's miner strategy of sending blocks to only 51% of the >> hashpower. > Speaking of, I hadn't gotten around to doing up the math behind that > strategy properly; turns out 51% I was overly optimistic and the actual > threshold is 29.3% > > Suppose I find a block. I have Q hashing power, and the rest of the > network 1-Q. Should I tell the rest of the network, or withhold that > block and hope I find a second one? > > Now in a purely inflation subsidy environment, where I don't care about > the other miners success, of course I should publish. However, if my > goals are to find *more* blocks than the other miners for whatever > reason, maybe because transaction fees matter or I'm trying to get > nLockTime'd announce/commit fee sacrifices, it gets more complicated. > > > There are three possible outcomes: > > 1) I find the next block, probability Q > 2) They find the next block, probability 1-Q > 2.1) I find the next block, probability Q, or (1-Q)*Q in total. > 2.2) They find the next block, probability (1-Q)^2 in total. > > Note how only in the last option do I lose. So how much hashing power do > I need before it is just as likely that the other miners will find two > blocks before I find either one block, or two blocks? Easy enough: > > Q + (1-Q)*Q = (1-Q)^2 -> Q^2 - Q + 1/2 -> Q = (1 - \sqrt(2))/2 > > Q ~= 29.2% > > So basically, if I'm trying to beat other miners, once I have >29.3% of > the hashing power I have no incentive to publish the blocks I mine! > > But hang on, does it matter if I'm the one who actually has that hashing > power? What if I just make sure that only >29.3% of the hashing power > has that block? If my goal is to make sure that someone does useless > work, and/or they are working on a lower height block than me, then no, > I don't care, which means my original "send blocks to >51% of the > hashing power" analysis was actually wrong, and the strategy is even > more crazy: "send blocks to >29.3% of the hashing power" (!) > > > Lets suppose I know that I'm two blocks ahead: > > 1) I find the next block: Q (3:0) > 2) They find the next block: (1-Q) (2:1) > 2.1) I find the next block: (1-Q)*Q (3:1) > 2.2) They find the next block: (1-Q)^2 (2:2) > 2.2.1) I find the next block: (1-Q)^2 * Q (3:2) > 2.2.2) They find the next block: (1-Q)^3 (2:3) > > At what hashing power should I release my blocks? So remember, I win > this round on outcomes 1, 2.1, 2.2.1 and they only win on 2.2.2: > > Q + (1-Q)*Q + (1-Q)^2*Q = (1-Q)^3 -> Q = 1 - 2^-3 > > Q ~= 20.6% > > Interesting... so as I get further ahead, or to be exact the group of > miners who have a given block gets further ahead, I need less hashing > power for my incentives to be to *not* publish the block I just found. > Conversely this means I should try to make my blocks propagate to less > of the hashing power, by whatever means necessary. > > Now remember, none of the above strategy requires me to have a special > low-latency network or anything fancy. I don't even have to have a lot > of hashing power - the strategy still works if I'm, say, a 5% pool. It > just means I don't have the incentives people thought I did to propagate > my blocks widely. > > The other nasty thing about this, is suppose I'm a miner and recently > got a block from another miner: should I forward that block, or not > bother? Well, it depends: if I have no idea how much of the hashing > power has that block, I should forward the block. But again, if my goal > is to be most likely to get the next block, I should only forward in > such a way that >30% of the hashing power has the block. > > This means that if I have some information about what % already has that > block, I have less incentive to forward! For instance, suppose that > every major miner has been publishing their node addresses in their > blocks - I'll have a pretty good idea of who probably has that most > recent block, so I can easily make a well-optimized decision not to > forward. Similarly because the 30% hashing power figure is the > *integral* of time * hashes/second, if miners are forwarding > near-target-headers, I might as well wait a few seconds and see if I see > any near-target-headers; if I do for this block then I have evidence > that hashing power does have it, and I shouldn't forward. > > > So yeah, we're fucked and have got to fix this awful incentive structure > somehow before the inflation subsidy gets any smaller. Also, raising the > blocksize, especially by just removing the limit, is utter madness given > it can be used to slow down block propagation selectively, so the > hashing power that gets a given block is limited repeatably to the same > group. > > > P.S: If any large pools want to try this stuff out, give me a shout. You > have my PGP key - confidentiality assured. > > P.P.S: If you're mining on a pool with more than, like, 1% hashing > power, do the math on varience... Seriously, stop it and go mine on a > smaller pool, or better yet, p2pool. > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > > > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development
On Wed, Nov 06, 2013 at 10:15:40PM -0600, Kyle Jerviss wrote:
Quoted text (3 lines)
> You are ignoring the gambler's ruin. We do not operate on an > infinite timeline. If you find a big pool willing to try this, > please give me enough advance warning to get my popcorn ready.
Gamblers ruin has nothing to do with it.
At every point you want to evaluate the chance the other side will get ahead, vs. cashing in by just publishing the blocks you have. (or some of them) I didn't mention it in the analysis, but obviously you want to keep track of how much the blocks you haven't published are worth to you, and consider publishing some or all of your lead to the rest of the network if you stand to lose more than you gain.
Right now it's a mostly theoretical attack because the inflation subsidy is enormous and fees don't matter, but once fees do start to matter things get a lot more complex. An extreme example is announce/commit sacrifices to mining fees: if I'm at block n+1, the rest of the network is at block n, and there's a 100BTC sacrifice at block n+2, I could easily be in a situation where I have zero incentive to publish my block to keep everyone else behind me, and just hope I find block n+2. If I do, great! I'll immediately publish to lock-in my winnings and start working on block n+3
Anyway, my covert suggestion that pools contact me was more to hopefully strike fear into the people mining at a large pool and get them to switch to a small one. :) If everyone mined solo or on p2pool none of this stuff would matter much... but we can't force them too yet.
Quoted text (3 lines)
> P.S: If any large pools want to try this stuff out, give me a shout. You > have my PGP key - confidentiality assured. >
If I find out one of the large pools decides to run this 'experiment' on the main network, I will make it my mission to tell people to switch to a more responsible pool.
And if you think you can get away with driving up EVERYBODY's orphan rate without anybody noticing, you should think again.
Quoted text (4 lines)
> P.P.S: If you're mining on a pool with more than, like, 1% hashing > power, do the math on varience... Seriously, stop it and go mine on a > smaller pool, or better yet, p2pool. >
That I agree with.
Each block that you solve has a reward. In practice, some blocks will be orphaned, so the expected reward is slightly less than the nominal reward. Each second that you delay publishing a block, the expected reward drops somewhat.
On an infinite timeline, the total reward approaches the expected reward. But reality is discrete, and zero tends to be a brick wall. If you delay publishing a block, you will get either the nominal reward, or zero, not some fraction in between. And if your personal random walk involves an excursion through negative land, you may not stick around long enough for it to come back.
Thus, a positive expected value is not sufficient for some strategy to be a good one.
Peter Todd wrote:
Quoted text (43 lines)
> On Wed, Nov 06, 2013 at 10:15:40PM -0600, Kyle Jerviss wrote: >> You are ignoring the gambler's ruin. We do not operate on an >> infinite timeline. If you find a big pool willing to try this, >> please give me enough advance warning to get my popcorn ready. > Gamblers ruin has nothing to do with it. > > At every point you want to evaluate the chance the other side will get > ahead, vs. cashing in by just publishing the blocks you have. (or some > of them) I didn't mention it in the analysis, but obviously you want to > keep track of how much the blocks you haven't published are worth to > you, and consider publishing some or all of your lead to the rest of the > network if you stand to lose more than you gain. > > Right now it's a mostly theoretical attack because the inflation subsidy > is enormous and fees don't matter, but once fees do start to matter > things get a lot more complex. An extreme example is announce/commit > sacrifices to mining fees: if I'm at block n+1, the rest of the network > is at block n, and there's a 100BTC sacrifice at block n+2, I could > easily be in a situation where I have zero incentive to publish my block > to keep everyone else behind me, and just hope I find block n+2. If I > do, great! I'll immediately publish to lock-in my winnings and start > working on block n+3 > > > Anyway, my covert suggestion that pools contact me was more to hopefully > strike fear into the people mining at a large pool and get them to > switch to a small one. :) If everyone mined solo or on p2pool none of > this stuff would matter much... but we can't force them too yet. > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > > > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development
What I want is configurable 1/10/100 millisecond ticks, and accurate flow of information.
It doesn't seem necessary to really emulate the whole protocol, nor to be overly concerned with the content of messages, nor to simulate every little housekeeping step or network message.
I'm not looking for a bitcoin-network-in-a-bottle, I just want to see flows. In the current situation, how often does a miner win if they hold their block until they see another one? How does that change with various numbers of remote sensors?
Other applications in the future could very well involve transaction spread, double spends, network partitions, transaction replacement, etc.
If the simulation run in question involves blocks, I'd like realistic latencies for blocks. If it is about transactions, the latencies should be realistic for transactions.
What is realistic for those? That brings me to...
I'll kick in another 1 BTC for an instrumentation package for the reference client. Same conditions as before. A runtime option, disabled by default, that collects data for the simulator. If this creates an uproar, I'll also accept a compile-time option. Support dumping to a file that can be uploaded to a parser as the bare minimum, and if you are feeling clever, add automatic uploads to a server specified in the conf file, or whatever. All data should be anonymous, of course. Local file should be in a format that humans can read (JSON, XML, CSV, etc) so that people can verify that the data is indeed anonymous.
I want stats on peers (number, turnover, latency, in/out, etc), stats on local operations (I/O stats, sigs per second when verifying a block, fraction of sig cache hits when validating, etc) and whatever else might be useful to a simulator. Each parameter should collect min, max, mean, std. deviation, etc so that the simulator can provide realistic virtual nodes.
Also, I don't want anyone to think that they need to satisfy me personally to collect on either of these two bounties. I will pay mine for a product that is generally along the lines I have laid out, if a couple of the core devs (Gavin, Greg, Jeff, sipa, Luke, etc) agree that your work is useful.
Christophe Biocca wrote:
Quoted text (66 lines)
> > I might try building this sometime soon. I think it may also serve an > educational purpose when trying to understand the whole network's > behaviour. > > What level of accuracy are we looking for though? Obviously we need to > fully emulate the steps of the network protocol, and we need to be > able to specify time taken for transmission/processing for each node. > Do we care about the actual contents of the messages (to be able to > simulate double spend attempts, invalid transactions and blocks, SPV > node communication), and their validation (actual signatures and proof > of work)? > > I imagine the latter is pretty useless, beyond specifying that the > signature/proof of work is valid/invalid. > > If we could build up a set of experiments we'd like to run on it, it > would help clarify what's needed. > > Off the top of my head: > > - Peter Todd's miner strategy of sending blocks to only 51% of the > hashpower. > - Various network split conditions, and how aware of the split nodes > would be (and the effect of client variability). > - Testing the feasability of network race double spends, or Finney > attacks. > - Various network partition scenarios. > - Tricking SPV nodes. > > On Nov 6, 2013 6:37 AM, "Jeff Garzik" <jgarzik@bitpay•com > <mailto:jgarzik@bitpay.com>> wrote: > > I will contribute 1 BTC to this bounty, under same terms and > expiration. > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming > models. Explore > techniques for threading, error checking, porting, and tuning. Get > the most > from the latest Intel processors and coprocessors. See abstracts > and register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > <mailto:Bitcoin-development@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/bitcoin-development > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > > > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development
I wonder if you need to take into consideration the fact that there might be another "bad" pool (in the 1-Q part of the network) running the same strategy and also holding on to two blocks of their own? Once they find their third block before you do, then your 2 blocks lead is gone instantly.
On Wed, Nov 06, 2013 at 10:59:28PM -0600, Kyle Jerviss wrote:
Quoted text (4 lines)
> Each block that you solve has a reward. In practice, some blocks > will be orphaned, so the expected reward is slightly less than the > nominal reward. Each second that you delay publishing a block, the > expected reward drops somewhat.
You don't understand how to read papers.
A good author will state his assumptions. For instance my third paragraph read:
Now in a purely inflation subsidy environment, where I don't care about
the other miners success, of course I should publish. However, if my
goals are to find *more* blocks than the other miners for whatever
reason, maybe because transaction fees matter or I'm trying to get
nLockTime'd announce/commit fee sacrifices, it gets more complicated.
Now that you understand the assumptions made, you can attack the paper in one of two ways:
1) Show it's wrong.
2) Show its assumptions make it irrelevant.
You've done neither.
On Thu, Nov 07, 2013 at 02:56:56PM +1000, Gavin Andresen wrote:
Quoted text (7 lines)
> > P.S: If any large pools want to try this stuff out, give me a shout. You > > have my PGP key - confidentiality assured. > > > > If I find out one of the large pools decides to run this 'experiment' on > the main network, I will make it my mission to tell people to switch to a > more responsible pool.
I hope they listen.
A few months ago ASICMiner could have made use of that attack if my memories of their peak hashing power were correct. They certainely could have used the selfish miner version, (we need better name for that) although development costs would eat into profits.
GHash.IO, 22%, says they're a "private Bitfury ASIC mining pool" - dunno what they mean by that, but they're involved with CEX.IO who has physical control of a bunch of hashing power so I guess that means their model is like ASICMiners. They're a bit short of 30%, but maybe some behind-the-scenes deals would fix that, and/or lowering the barrier with reactive block publishing. (a better name)
Quoted text (2 lines)
> And if you think you can get away with driving up EVERYBODY's orphan rate > without anybody noticing, you should think again.
...and remember, if you only do the attack a little bit, you still can earn more profit, and only drive up the orphan rate a little bit. So who knows, maybe the orphans are real, or maybe they're an attack? ASICMiner was involved with a bunch of orphans a while back...
You know what this calls for? A witchhunt!
BURN THE LARGE POOLS!
Quoted text (6 lines)
> > P.P.S: If you're mining on a pool with more than, like, 1% hashing > > power, do the math on varience... Seriously, stop it and go mine on a > > smaller pool, or better yet, p2pool. > > > > That I agree with.
Glad to hear.
Once the ASIC race calms down because everyone has one, has more or less optimal power supplies, process improvements aren't easily reachable anymore etc then I'd expect people to dissipate from the large pools because eliminating their fees will become the next lowest hanging fruit to squeeze out extra profit. There's no particular reason we need only a handful of pools that control a major fraction of the hashpower.
If we end up with a few hundred pools or lots of miners on p2pool, then a lot of these theoretical attacks become not very relevant (I don't think ID sacrifices will be so common or large as to justify a pile of custom mining code+strategies at any point ...)
On Thu, Nov 7, 2013 at 2:24 PM, Peter Todd <pete@petertodd•org> wrote:
Quoted text (64 lines)
> On Thu, Nov 07, 2013 at 02:56:56PM +1000, Gavin Andresen wrote: > > > P.S: If any large pools want to try this stuff out, give me a shout. > You > > > have my PGP key - confidentiality assured. > > > > > > > If I find out one of the large pools decides to run this 'experiment' on > > the main network, I will make it my mission to tell people to switch to a > > more responsible pool. > > I hope they listen. > > A few months ago ASICMiner could have made use of that attack if my > memories of their peak hashing power were correct. They certainely could > have used the selfish miner version, (we need better name for that) > although development costs would eat into profits. > > GHash.IO, 22%, says they're a "private Bitfury ASIC mining pool" - dunno > what they mean by that, but they're involved with CEX.IO who has > physical control of a bunch of hashing power so I guess that means their > model is like ASICMiners. They're a bit short of 30%, but maybe some > behind-the-scenes deals would fix that, and/or lowering the barrier with > reactive block publishing. (a better name) > > > And if you think you can get away with driving up EVERYBODY's orphan rate > > without anybody noticing, you should think again. > > ...and remember, if you only do the attack a little bit, you still can > earn more profit, and only drive up the orphan rate a little bit. So who > knows, maybe the orphans are real, or maybe they're an attack? ASICMiner > was involved with a bunch of orphans a while back... > > You know what this calls for? A witchhunt! > > BURN THE LARGE POOLS! > > > > P.P.S: If you're mining on a pool with more than, like, 1% hashing > > > power, do the math on varience... Seriously, stop it and go mine on a > > > smaller pool, or better yet, p2pool. > > > > > > > That I agree with. > > Glad to hear. > > -- > 'peter'[:-1]@petertodd.org > 0000000000000007bd936f19e33bc8b8f9bb1f4c013b863ef60a7f5a6a5d2112 > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development > >
Hey Peter, something seems wrong with your above analysis: I think a miner would withhold his block not because it leads to a greater probability of winning the next one, but because it increases his expected revenue.
Suppose a cabal with fraction q of the total hashing power is n blocks ahead on a secret branch of that has mined r_tot coins, and let r_next be its next block's reward. If the cabal chooses not to broadcast its secret chain until at least the next block, its expected revenue after the next block is found is
(1 - (1-q)^(n+1))*(r_tot + r_next)
If it does broadcast, its expected revenue after the next block is found is
r_tot + q * r_next
If the cabal seeks only to maximize immediate revenue, then after a bit of algebra we find that it will withhold its chain if
q > 1 - ( 1 + r_tot / r_next )^(-1/n)
So if the cabal has just mined his first block off of the public chain, i.e. n = 1, and if the block reward is relatively stable, i.e. r_next = r_tot, then it needs q > 50% to profitably withhold, not the 29.2% you calculated.
From this formula we can also see that if the miner wins the race and withholds again, then he must grow q to compensate for the increase in r_tot, and any decrease in n. So generally publication becomes increasingly in the cabal's interest, and secret chains will tend not to grow too large (intuition tells me that simulations using the above formula should bear this out).
This seem correct to you?
On Thu, Nov 7, 2013 at 9:14 AM, Mike Hearn <mike@plan99•net> wrote:
Quoted text (99 lines)
> Once the ASIC race calms down because everyone has one, has more or less > optimal power supplies, process improvements aren't easily reachable > anymore etc then I'd expect people to dissipate from the large pools > because eliminating their fees will become the next lowest hanging fruit to > squeeze out extra profit. There's no particular reason we need only a > handful of pools that control a major fraction of the hashpower. > > If we end up with a few hundred pools or lots of miners on p2pool, then a > lot of these theoretical attacks become not very relevant (I don't think ID > sacrifices will be so common or large as to justify a pile of custom mining > code+strategies at any point ...) > > > On Thu, Nov 7, 2013 at 2:24 PM, Peter Todd <pete@petertodd•org> wrote: > >> On Thu, Nov 07, 2013 at 02:56:56PM +1000, Gavin Andresen wrote: >> > > P.S: If any large pools want to try this stuff out, give me a shout. >> You >> > > have my PGP key - confidentiality assured. >> > > >> > >> > If I find out one of the large pools decides to run this 'experiment' on >> > the main network, I will make it my mission to tell people to switch to >> a >> > more responsible pool. >> >> I hope they listen. >> >> A few months ago ASICMiner could have made use of that attack if my >> memories of their peak hashing power were correct. They certainely could >> have used the selfish miner version, (we need better name for that) >> although development costs would eat into profits. >> >> GHash.IO, 22%, says they're a "private Bitfury ASIC mining pool" - dunno >> what they mean by that, but they're involved with CEX.IO who has >> physical control of a bunch of hashing power so I guess that means their >> model is like ASICMiners. They're a bit short of 30%, but maybe some >> behind-the-scenes deals would fix that, and/or lowering the barrier with >> reactive block publishing. (a better name) >> >> > And if you think you can get away with driving up EVERYBODY's orphan >> rate >> > without anybody noticing, you should think again. >> >> ...and remember, if you only do the attack a little bit, you still can >> earn more profit, and only drive up the orphan rate a little bit. So who >> knows, maybe the orphans are real, or maybe they're an attack? ASICMiner >> was involved with a bunch of orphans a while back... >> >> You know what this calls for? A witchhunt! >> >> BURN THE LARGE POOLS! >> >> > > P.P.S: If you're mining on a pool with more than, like, 1% hashing >> > > power, do the math on varience... Seriously, stop it and go mine on a >> > > smaller pool, or better yet, p2pool. >> > > >> > >> > That I agree with. >> >> Glad to hear. >> >> -- >> 'peter'[:-1]@petertodd.org >> 0000000000000007bd936f19e33bc8b8f9bb1f4c013b863ef60a7f5a6a5d2112 >> >> >> ------------------------------------------------------------------------------ >> November Webinars for C, C++, Fortran Developers >> Accelerate application performance with scalable programming models. >> Explore >> techniques for threading, error checking, porting, and tuning. Get the >> most >> from the latest Intel processors and coprocessors. See abstracts and >> register >> >> http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk >> _______________________________________________ >> Bitcoin-development mailing list >> Bitcoin-development@lists•sourceforge•net >> https://lists.sourceforge.net/lists/listinfo/bitcoin-development >> >> > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development > >
Nicholas Weaver is reporting that pools have already started delaying blocks, something that hints at Selfish Mining, since Nov. 3rd. https://medium.com/something-like-falling/d321a2ef9317
He dismisses other reasons for delayed block propagation.
Any ideas on whether pools are already mucking around with block delaying tactics?
I have no idea if this report is accurate or explained by some other issue in the network, does anyone here have a comment on this?
On Thu, Nov 7, 2013 at 10:28 AM, Daniel Lidstrom <lidstrom83@gmail•com>wrote:
Quoted text (156 lines)
> Hey Peter, something seems wrong with your above analysis: I think a miner > would withhold his block not because it leads to a greater probability of > winning the next one, but because it increases his expected revenue. > > Suppose a cabal with fraction q of the total hashing power is n blocks > ahead on a secret branch of that has mined r_tot coins, and let r_next be > its next block's reward. If the cabal chooses not to broadcast its secret > chain until at least the next block, its expected revenue after the next > block is found is > > (1 - (1-q)^(n+1))*(r_tot + r_next) > > If it does broadcast, its expected revenue after the next block is found is > > r_tot + q * r_next > > If the cabal seeks only to maximize immediate revenue, then after a bit of > algebra we find that it will withhold its chain if > > q > 1 - ( 1 + r_tot / r_next )^(-1/n) > > So if the cabal has just mined his first block off of the public chain, > i.e. n = 1, and if the block reward is relatively stable, i.e. r_next = > r_tot, then it needs q > 50% to profitably withhold, not the 29.2% you > calculated. > > From this formula we can also see that if the miner wins the race and > withholds again, then he must grow q to compensate for the increase in > r_tot, and any decrease in n. So generally publication becomes > increasingly in the cabal's interest, and secret chains will tend not to > grow too large (intuition tells me that simulations using the above formula > should bear this out). > > This seem correct to you? > > > On Thu, Nov 7, 2013 at 9:14 AM, Mike Hearn <mike@plan99•net> wrote: > >> Once the ASIC race calms down because everyone has one, has more or less >> optimal power supplies, process improvements aren't easily reachable >> anymore etc then I'd expect people to dissipate from the large pools >> because eliminating their fees will become the next lowest hanging fruit to >> squeeze out extra profit. There's no particular reason we need only a >> handful of pools that control a major fraction of the hashpower. >> >> If we end up with a few hundred pools or lots of miners on p2pool, then a >> lot of these theoretical attacks become not very relevant (I don't think ID >> sacrifices will be so common or large as to justify a pile of custom mining >> code+strategies at any point ...) >> >> >> On Thu, Nov 7, 2013 at 2:24 PM, Peter Todd <pete@petertodd•org> wrote: >> >>> On Thu, Nov 07, 2013 at 02:56:56PM +1000, Gavin Andresen wrote: >>> > > P.S: If any large pools want to try this stuff out, give me a shout. >>> You >>> > > have my PGP key - confidentiality assured. >>> > > >>> > >>> > If I find out one of the large pools decides to run this 'experiment' >>> on >>> > the main network, I will make it my mission to tell people to switch >>> to a >>> > more responsible pool. >>> >>> I hope they listen. >>> >>> A few months ago ASICMiner could have made use of that attack if my >>> memories of their peak hashing power were correct. They certainely could >>> have used the selfish miner version, (we need better name for that) >>> although development costs would eat into profits. >>> >>> GHash.IO, 22%, says they're a "private Bitfury ASIC mining pool" - dunno >>> what they mean by that, but they're involved with CEX.IO who has >>> physical control of a bunch of hashing power so I guess that means their >>> model is like ASICMiners. They're a bit short of 30%, but maybe some >>> behind-the-scenes deals would fix that, and/or lowering the barrier with >>> reactive block publishing. (a better name) >>> >>> > And if you think you can get away with driving up EVERYBODY's orphan >>> rate >>> > without anybody noticing, you should think again. >>> >>> ...and remember, if you only do the attack a little bit, you still can >>> earn more profit, and only drive up the orphan rate a little bit. So who >>> knows, maybe the orphans are real, or maybe they're an attack? ASICMiner >>> was involved with a bunch of orphans a while back... >>> >>> You know what this calls for? A witchhunt! >>> >>> BURN THE LARGE POOLS! >>> >>> > > P.P.S: If you're mining on a pool with more than, like, 1% hashing >>> > > power, do the math on varience... Seriously, stop it and go mine on a >>> > > smaller pool, or better yet, p2pool. >>> > > >>> > >>> > That I agree with. >>> >>> Glad to hear. >>> >>> -- >>> 'peter'[:-1]@petertodd.org >>> 0000000000000007bd936f19e33bc8b8f9bb1f4c013b863ef60a7f5a6a5d2112 >>> >>> >>> ------------------------------------------------------------------------------ >>> November Webinars for C, C++, Fortran Developers >>> Accelerate application performance with scalable programming models. >>> Explore >>> techniques for threading, error checking, porting, and tuning. Get the >>> most >>> from the latest Intel processors and coprocessors. See abstracts and >>> register >>> >>> http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk >>> _______________________________________________ >>> Bitcoin-development mailing list >>> Bitcoin-development@lists•sourceforge•net >>> https://lists.sourceforge.net/lists/listinfo/bitcoin-development >>> >>> >> >> >> ------------------------------------------------------------------------------ >> November Webinars for C, C++, Fortran Developers >> Accelerate application performance with scalable programming models. >> Explore >> techniques for threading, error checking, porting, and tuning. Get the >> most >> from the latest Intel processors and coprocessors. See abstracts and >> register >> >> http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk >> _______________________________________________ >> Bitcoin-development mailing list >> Bitcoin-development@lists•sourceforge•net >> https://lists.sourceforge.net/lists/listinfo/bitcoin-development >> >> > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development > >
On Fri, Nov 8, 2013 at 11:49 AM, Andreas M. Antonopoulos <andreas@rooteleven•com> wrote:
Quoted text (11 lines)
> Nicholas Weaver is reporting that pools have already started delaying > blocks, something that hints at Selfish Mining, since Nov. 3rd. > https://medium.com/something-like-falling/d321a2ef9317 > > He dismisses other reasons for delayed block propagation. > > Any ideas on whether pools are already mucking around with block delaying > tactics? > > I have no idea if this report is accurate or explained by some other issue > in the network, does anyone here have a comment on this?
The BC.i timestamps have historically been inaccurate relative to my local GPS clock measurements on my own nodes... but not just that, it sounds like he's comparing block timestamps and bc.i numbers.
Thats insane, because it tells you the delay between when the miner _started_ a work unit and when BC.i claims to have found it. Even assuming bc.i's times were accurate and assuming miner clocks are accurate (they are often not) you expect there to be be a gap because it takes time to compute work, send it to the miner, search for a valid nonce (an average of 2^31 hash operations, often executed sequentially on a single core taking ten seconds or so on a lot of hardware) and then return a result.
Evidence of selfish miners wouldn't be block timestamps (which are inaccurate and controlled by miners anyways), or data on blockchain.info (which is inaccurate and controlled by bc.i) ... but the existence of an unusual amount of orphan blocks. High levels of blocks are _necessary_ evidence of this sort of things, there can be other explanations of high orphaning levels, but they're required here and couldn't be faked.
On Thu, Nov 07, 2013 at 11:28:52AM -0700, Daniel Lidstrom wrote:
Quoted text (34 lines)
> Hey Peter, something seems wrong with your above analysis: I think a miner > would withhold his block not because it leads to a greater probability of > winning the next one, but because it increases his expected revenue. > > Suppose a cabal with fraction q of the total hashing power is n blocks > ahead on a secret branch of that has mined r_tot coins, and let r_next be > its next block's reward. If the cabal chooses not to broadcast its secret > chain until at least the next block, its expected revenue after the next > block is found is > > (1 - (1-q)^(n+1))*(r_tot + r_next) > > If it does broadcast, its expected revenue after the next block is found is > > r_tot + q * r_next > > If the cabal seeks only to maximize immediate revenue, then after a bit of > algebra we find that it will withhold its chain if > > q > 1 - ( 1 + r_tot / r_next )^(-1/n) > > So if the cabal has just mined his first block off of the public chain, > i.e. n = 1, and if the block reward is relatively stable, i.e. r_next = > r_tot, then it needs q > 50% to profitably withhold, not the 29.2% you > calculated. > > From this formula we can also see that if the miner wins the race and > withholds again, then he must grow q to compensate for the increase in > r_tot, and any decrease in n. So generally publication becomes > increasingly in the cabal's interest, and secret chains will tend not to > grow too large (intuition tells me that simulations using the above formula > should bear this out). > > This seem correct to you?
Remember how I started off by asking what was the correct strategy if a miner wanted to get more blocks than their *competition*, not more blocks in total. In some scenarios that strategy is the one that maximizes returns, such as the case when you make your returns from transaction fees, especially without a blocksize limit restricting how many fee paying transactions you can stuff in your blocks. It's not correct to say the cabal is trying to maximize immediate revenue.
As for the length of those secret chains, at every step you of course want to weigh the value of the blocks you have found against the risk that someone else catches up, and when it makes sense, publish some or all.
I've created a simulation framework called simbit to simulate the selfish mining attack, though it is general enough to simulate any p2p network. I even put together a rough simulation of MinCen. The goal is to be fun/easy to rapidly prototype protocols and strategies, and visualize them. It's written in javascript, so it can be demoed in the web browser or run on Node.
It's still in early alpha and a lot of things are missing.
https://github.com/ebfull/simbit
https://bitcointalk.org/index.php?topic=603171.0
Feedback is appreciated!
On Tue, Nov 5, 2013 at 10:33 PM, kjj <bitcoin-devel@jerviss•org> wrote:
Quoted text (59 lines)
> One of the things that really gets me going is when someone devises a > model, tests it against itself, and then pretends that they've learned > something about the real world. > > Naturally, the Selfish Mining paper is exactly this sort of nonsense. > Their model is one with no latency, and one where the attacker has total > visibility across the network. An iterated FSM is not a suitable > simulation of the bitcoin system. The bitcoin network does not have > states, and to the extent that you can pretend that we do, you can't > simulate transitions between them with static probabilities. > > The authors understand this deep down inside, even though they didn't > work out the implications. They handwave the issue by assuming a total > sybil attack, and in true academic spirit, they don't realize that the > condition necessary for the attack is far, far worse than the attack > itself. > > Greg said he'd like to run some simulations, and I'm thinking about it > too. Unfortunately, he is busy all week, and I'm lazy (and also busy > for most of tomorrow). > > If neither of us get to it first, I'm willing to pitch in 1 BTC as a > bounty for building a general bitcoin network simulator framework. The > simulator should be able to account for latency between nodes, and > ideally within a node. It needs to be able to simulate an attacker that > owns varying fractions of the network, and make decisions based only on > what the attacker actually knows. It needs to be able to simulate this > "attack" and should be generic enough to be easily modified for other > crazy schemes. > > (Bounty offer is serious, but expires in one year [based on the earliest > timestamp that my mail server puts on this email], and /may/ be subject > to change if the price on any reputable exchange breaks 1000 USD per BTC > in that period.) > > Basically, the lack of a decent network simulator is what allowed this > paper to get press. If the author had been able to see the importance > of the stuff he was ignoring, we wouldn't be wasting so much time > correcting him (and sadly the reporters that have no way to check his > claims). > > https://bitcointalk.org/index.php?topic=324413.msg3495663#msg3495663 > > > > > ------------------------------------------------------------------------------ > November Webinars for C, C++, Fortran Developers > Accelerate application performance with scalable programming models. > Explore > techniques for threading, error checking, porting, and tuning. Get the most > from the latest Intel processors and coprocessors. See abstracts and > register > http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development >