Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
On Mon, Oct 27, 2014 at 03:33:45PM -0400, Alex Morcos wrote: I've been playing around with the code for estimating fees and found a few issues with the existing code. I think this will address several observations that the estimates returned by the existing code appear to be too high. For instance see @cozz in Issue 4866 https://github.com/bitcoin/bitcoin/issues/4866. I don't have time to look at the details of your statistical methods unfortunately due to some deadlines, but a quick comment: You should think about the malleability of your estimates to attackers. For instance the current fee estimation code has a serious issue where it'll happily estimate ludicriously high fees based on very little date. There is a 'insane fees' failsafe, but it's IIRC set to allow transactions with fees of less than 100mBTC/tx, roughly $50 at current exchange rates. It's relatively easy to get a wallet into a condition where this happens as the estimations are considered valid even based on very little data - a simple sybil attack suffices. (e.g. the recently published paper(1) on Tor sybil attacks comes to mind as one example of many ways to do this) Obviously this could empty someone's wallet pretty quickly; an exchange that makes a few dozen transactions an hour could easily lose tens of thousands of dollars due to this exploit. Someone correct me if I'm wrong, but last I checked in git HEAD this exploit is still unfixed. A user-configurable failsafe limit is a pretty obvious solution here, albeit a crude one; it'd be interesting to see if a plausible security argument could be made for something more sophisticated, like taking into account coin-age of observed transactions that estimates are based on. 1) Bitcoin over Tor isn't a good idea, http://arxiv.org/abs/1410.6079 -- 'peter'[:-1]@petertodd.org 098d3c9095b47ff1fd692fef5ac6731340802c7c63d38bb0 signature.asc Description: Digital signature -- ___ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
Could you explain a little further why you think the current approach is statistically incorrect? There's no doubt that the existing estimates the system produces are garbage, but that's because it assumes players in the fee market are rational and they are not. Fwiw bitcoinj 0.12.1 applies the January fee drop and will attach fee of only 1000 satoshis per kB by default. I also have a program that measures confirmation time for a given fee level (with fresh coins so there's no priority) and it aligns with your findings, most txns confirm within a couple of blocks. Ultimately there isn't any easy method to stop people throwing money away. Bitcoinj will probably continue to use hard coded fee values for now to try and contribute to market sanity in the hope it makes smartfees smarter. On 27 Oct 2014 19:34, Alex Morcos mor...@gmail.com wrote: I've been playing around with the code for estimating fees and found a few issues with the existing code. I think this will address several observations that the estimates returned by the existing code appear to be too high. For instance see @cozz in Issue 4866 https://github.com/bitcoin/bitcoin/issues/4866. Here's what I found: 1) We're trying to answer the question of what fee X you need in order to be confirmed within Y blocks. The existing code tries to do that by calculating the median fee for each possible Y instead of gathering statistics for each possible X. That approach is statistically incorrect. In fact since certain X's appear so frequently, they tend to dominate the statistics at all possible Y's (a fee rate of about 40k satoshis) 2) The existing code then sorts all of the data points in all of the buckets together by fee rate and then reassigns buckets before calculating the medians for each confirmation bucket. The sorting forces a relationship where there might not be one. Imagine some other variable, such as first 2 bytes of the transaction hash. If we sorted these and then used them to give estimates, we'd see a clear but false relationship where transactions with low starting bytes in their hashes took longer to confirm. 3) Transactions which don't have all their inputs available (because they depend on other transactions in the mempool) aren't excluded from the calculations. This skews the results. I rewrote the code to follow a different approach. I divided all possible fee rates up into fee rate buckets (I spaced these logarithmically). For each transaction that was confirmed, I updated the appropriate fee rate bucket with how many blocks it took to confirm that transaction. The hardest part of doing this fee estimation is to decide what the question really is that we're trying to answer. I took the approach that if you are asking what fee rate I need to be confirmed within Y blocks, then what you would like to know is the lowest fee rate such that a relatively high percentage of transactions of that fee rate are confirmed within Y blocks. Since even the highest fee transactions are confirmed within the first block only 90-93% of the time, I decided to use 80% as my cutoff. So now to answer estimatefee Y, I scan through all of the fee buckets from the most expensive down until I find the last bucket with 80% of the transactions confirmed within Y blocks. Unfortunately we still have the problem of not having enough data points for non-typical fee rates, and so it requires gathering a lot of data to give reasonable answers. To keep all of these data points in a circular buffer and then sort them for every analysis (or after every new block) is expensive. So instead I adopted the approach of keeping an exponentially decaying moving average for each bucket. I used a decay of .998 which represents a half life of 374 blocks or about 2.5 days. Also if a bucket doesn't have very many transactions, I combine it with the next bucket. Here is a link https://github.com/morcos/bitcoin/pull/3 to the code. I can create an actual pull request if there is consensus that it makes sense to do so. I've attached a graph comparing the estimates produced for 1-3 confirmations by the new code and the old code. I did apply the patch to fix issue 3 above to the old code first. The new code is in green and the fixed code is in purple. The Y axis is a log scale of feerate in satoshis per KB and the X axis is chain height. The new code produces the same estimates for 2 and 3 confirmations (the answers are effectively quantized by bucket). I've also completely reworked smartfees.py. It turns out to require many many more transactions are put through in order to have statistically significant results, so the test is quite slow to run (about 3 mins on my machine). I've also been running a real world test, sending transactions of various fee rates and seeing how long they took to get confirmed. After almost 200 tx's at each fee rate, here are the results so far: Fee rate 1100
Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
I think Alex's approach is better; I don't think we can know how much better until we have a functioning fee market. We don't have a functioning fee market now, because fees are hard-coded. So we get pay the hard-coded fee and you'll get confirmed in one or two or three blocks, depending on which miners mine the next three blocks and what time of day it is. git HEAD code says you need a fee of 10, satoshis/kb to be pretty sure you'll get confirmed in the next block. That looks about right with Alex's real-world data (if we take 90% chance as 'pretty sure you'll get confirmed'): Fee rate 10 Avg blocks to confirm 1.09 NumBlocks:% confirmed 1: 0.901 2: 1.0 3: 1.0 My only concern with Alex's code is that it takes much longer to get 'primed' -- Alex, if I started with no data about fees, how long would it take to be able to get enough data for a reasonable estimate of what is the least I can pay and still be 90% sure I get confirmed in 20 blocks ? Hours? Days? Weeks? -- -- Gavin Andresen -- ___ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
Oh in just a couple of blocks, it'll give you a somewhat reasonable estimate for asking about every confirmation count other than 1, but it could take several hours for it to have enough data points to give you a good estimate for getting confirmed in one block (because the prevalent feerate is not always confirmed in 1 block 80% of the time) Essentially what it does is just combine buckets until it has enough data points, so after the first block it might be treating all of the txs as belonging to the same feerate bucket, but since the answer it returns is the median* fee rate for that bucket, its a reasonable answer right off the get go. Do you think it would make sense to make that 90% number an argument to rpc call? For instance there could be a default (I would use 80%) but then you could specify if you required a different certainty. It wouldn't require any code changes and might make it easier for people to build more complicated logic on top of it. *It can't actually track the median, but it identifies which of the smaller actual buckets the median would have fallen into and returns the average feerate for that median bucket. On Tue, Oct 28, 2014 at 9:59 AM, Gavin Andresen gavinandre...@gmail.com wrote: I think Alex's approach is better; I don't think we can know how much better until we have a functioning fee market. We don't have a functioning fee market now, because fees are hard-coded. So we get pay the hard-coded fee and you'll get confirmed in one or two or three blocks, depending on which miners mine the next three blocks and what time of day it is. git HEAD code says you need a fee of 10, satoshis/kb to be pretty sure you'll get confirmed in the next block. That looks about right with Alex's real-world data (if we take 90% chance as 'pretty sure you'll get confirmed'): Fee rate 10 Avg blocks to confirm 1.09 NumBlocks:% confirmed 1: 0.901 2: 1.0 3: 1.0 My only concern with Alex's code is that it takes much longer to get 'primed' -- Alex, if I started with no data about fees, how long would it take to be able to get enough data for a reasonable estimate of what is the least I can pay and still be 90% sure I get confirmed in 20 blocks ? Hours? Days? Weeks? -- -- Gavin Andresen -- ___ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
Sorry, perhaps I misinterpreted that question. The estimates will be dominated by the prevailing transaction rates initially, so the estimates you get for something like what is the least I can pay and still be 90% sure I get confirmed in 20 blocks won't be insane, but they will still be way too conservative. I'm not sure what you meant by reasonable. You won't get the correct answer of something significantly less than 40k sat/kB for quite some time. Given that the half-life of the decay is 2.5 days, then within a couple of days. And in fact even in the steady state, the new code will still return a much higher rate than the existing code, say 10k sat/kB instead of 1k sat/kB, but that's just a result of the sorting the existing code does and the fact that no one places transactions with that small fee. To correctly give such low answers, the new code will require that those super low feerate transactions are occurring frequently enough, but the bar for enough datapoints in a feerate bucket is pretty low, an average of 1 tx per block. The bar can be made lower at the expense of a bit of noisiness in the answers, for instance for priorities I had to make the bar significantly lower because there are so many fewer transactions confirmed because of priorities. I'm certainly open to tuning some of these variables. On Tue, Oct 28, 2014 at 10:30 AM, Alex Morcos mor...@gmail.com wrote: Oh in just a couple of blocks, it'll give you a somewhat reasonable estimate for asking about every confirmation count other than 1, but it could take several hours for it to have enough data points to give you a good estimate for getting confirmed in one block (because the prevalent feerate is not always confirmed in 1 block 80% of the time) Essentially what it does is just combine buckets until it has enough data points, so after the first block it might be treating all of the txs as belonging to the same feerate bucket, but since the answer it returns is the median* fee rate for that bucket, its a reasonable answer right off the get go. Do you think it would make sense to make that 90% number an argument to rpc call? For instance there could be a default (I would use 80%) but then you could specify if you required a different certainty. It wouldn't require any code changes and might make it easier for people to build more complicated logic on top of it. *It can't actually track the median, but it identifies which of the smaller actual buckets the median would have fallen into and returns the average feerate for that median bucket. On Tue, Oct 28, 2014 at 9:59 AM, Gavin Andresen gavinandre...@gmail.com wrote: I think Alex's approach is better; I don't think we can know how much better until we have a functioning fee market. We don't have a functioning fee market now, because fees are hard-coded. So we get pay the hard-coded fee and you'll get confirmed in one or two or three blocks, depending on which miners mine the next three blocks and what time of day it is. git HEAD code says you need a fee of 10, satoshis/kb to be pretty sure you'll get confirmed in the next block. That looks about right with Alex's real-world data (if we take 90% chance as 'pretty sure you'll get confirmed'): Fee rate 10 Avg blocks to confirm 1.09 NumBlocks:% confirmed 1: 0.901 2: 1.0 3: 1.0 My only concern with Alex's code is that it takes much longer to get 'primed' -- Alex, if I started with no data about fees, how long would it take to be able to get enough data for a reasonable estimate of what is the least I can pay and still be 90% sure I get confirmed in 20 blocks ? Hours? Days? Weeks? -- -- Gavin Andresen -- ___ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
On Tue, Oct 28, 2014 at 10:30 AM, Alex Morcos mor...@gmail.com wrote: Do you think it would make sense to make that 90% number an argument to rpc call? For instance there could be a default (I would use 80%) but then you could specify if you required a different certainty. It wouldn't require any code changes and might make it easier for people to build more complicated logic on top of it. RE: 80% versus 90% : I think a default of 80% will get us a lot of the fee estimation logic is broken, I want my transactions to confirm quick and a lot of them aren't confirming for 2 or 3 blocks. RE: RPC argument: I'm reluctant to give too many 'knobs' for the RPC interface. I think the default percentage makes sense as a command-line/bitcoin.conf option; I can imagine services that want to save on fees running with -estimatefeethreshold=0.5 (or -estimatefeethreshold=0.95 if as-fast-as-possible confirmations are needed). Setting both the number of confirmations and the estimation threshold on a transaction-by-transaction basis seems like overkill to me. -- -- Gavin Andresen -- ___ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
Re: [Bitcoin-development] Reworking the policy estimation code (fee estimates)
RE: 90% : I think it's fine to use 90% for anything other than 1 confirmation, but if you look at the real world data test I did, or the raw data from this new code, you'll see that even the highest fee rate transactions only get confirmed at about a 90% rate in 1 block, so that if you use that as your cut-off you will sometimes get no answer and sometimes get a very high fee rate and sometimes get a reasonable fee rate, it just depends because the data is too noisy. I think thats just because there is no good answer to that question. There is no fee you can put on your transaction to guarantee greater than 90% chance of getting confirmed in one block. I think 85% might be safe? RE: tunable as command-line/bitcoin.conf: sounds good! OK, sorry to have all this conversation on the dev list, maybe i'll turn this into an actual PR if we want to comment on the code? I just wanted to see if it even made sense to make a PR for this or this isn't the way we wanted to go about it. On Tue, Oct 28, 2014 at 10:58 AM, Gavin Andresen gavinandre...@gmail.com wrote: On Tue, Oct 28, 2014 at 10:30 AM, Alex Morcos mor...@gmail.com wrote: Do you think it would make sense to make that 90% number an argument to rpc call? For instance there could be a default (I would use 80%) but then you could specify if you required a different certainty. It wouldn't require any code changes and might make it easier for people to build more complicated logic on top of it. RE: 80% versus 90% : I think a default of 80% will get us a lot of the fee estimation logic is broken, I want my transactions to confirm quick and a lot of them aren't confirming for 2 or 3 blocks. RE: RPC argument: I'm reluctant to give too many 'knobs' for the RPC interface. I think the default percentage makes sense as a command-line/bitcoin.conf option; I can imagine services that want to save on fees running with -estimatefeethreshold=0.5 (or -estimatefeethreshold=0.95 if as-fast-as-possible confirmations are needed). Setting both the number of confirmations and the estimation threshold on a transaction-by-transaction basis seems like overkill to me. -- -- Gavin Andresen -- ___ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
[Bitcoin-development] Reworking the policy estimation code (fee estimates)
I've been playing around with the code for estimating fees and found a few issues with the existing code. I think this will address several observations that the estimates returned by the existing code appear to be too high. For instance see @cozz in Issue 4866 https://github.com/bitcoin/bitcoin/issues/4866. Here's what I found: 1) We're trying to answer the question of what fee X you need in order to be confirmed within Y blocks. The existing code tries to do that by calculating the median fee for each possible Y instead of gathering statistics for each possible X. That approach is statistically incorrect. In fact since certain X's appear so frequently, they tend to dominate the statistics at all possible Y's (a fee rate of about 40k satoshis) 2) The existing code then sorts all of the data points in all of the buckets together by fee rate and then reassigns buckets before calculating the medians for each confirmation bucket. The sorting forces a relationship where there might not be one. Imagine some other variable, such as first 2 bytes of the transaction hash. If we sorted these and then used them to give estimates, we'd see a clear but false relationship where transactions with low starting bytes in their hashes took longer to confirm. 3) Transactions which don't have all their inputs available (because they depend on other transactions in the mempool) aren't excluded from the calculations. This skews the results. I rewrote the code to follow a different approach. I divided all possible fee rates up into fee rate buckets (I spaced these logarithmically). For each transaction that was confirmed, I updated the appropriate fee rate bucket with how many blocks it took to confirm that transaction. The hardest part of doing this fee estimation is to decide what the question really is that we're trying to answer. I took the approach that if you are asking what fee rate I need to be confirmed within Y blocks, then what you would like to know is the lowest fee rate such that a relatively high percentage of transactions of that fee rate are confirmed within Y blocks. Since even the highest fee transactions are confirmed within the first block only 90-93% of the time, I decided to use 80% as my cutoff. So now to answer estimatefee Y, I scan through all of the fee buckets from the most expensive down until I find the last bucket with 80% of the transactions confirmed within Y blocks. Unfortunately we still have the problem of not having enough data points for non-typical fee rates, and so it requires gathering a lot of data to give reasonable answers. To keep all of these data points in a circular buffer and then sort them for every analysis (or after every new block) is expensive. So instead I adopted the approach of keeping an exponentially decaying moving average for each bucket. I used a decay of .998 which represents a half life of 374 blocks or about 2.5 days. Also if a bucket doesn't have very many transactions, I combine it with the next bucket. Here is a link https://github.com/morcos/bitcoin/pull/3 to the code. I can create an actual pull request if there is consensus that it makes sense to do so. I've attached a graph comparing the estimates produced for 1-3 confirmations by the new code and the old code. I did apply the patch to fix issue 3 above to the old code first. The new code is in green and the fixed code is in purple. The Y axis is a log scale of feerate in satoshis per KB and the X axis is chain height. The new code produces the same estimates for 2 and 3 confirmations (the answers are effectively quantized by bucket). I've also completely reworked smartfees.py. It turns out to require many many more transactions are put through in order to have statistically significant results, so the test is quite slow to run (about 3 mins on my machine). I've also been running a real world test, sending transactions of various fee rates and seeing how long they took to get confirmed. After almost 200 tx's at each fee rate, here are the results so far: Fee rate 1100 Avg blocks to confirm 2.30 NumBlocks:% confirmed 1: 0.528 2: 0.751 3: 0.870 Fee rate 2500 Avg blocks to confirm 2.22 NumBlocks:% confirmed 1: 0.528 2: 0.766 3: 0.880 Fee rate 5000 Avg blocks to confirm 1.93 NumBlocks:% confirmed 1: 0.528 2: 0.782 3: 0.891 Fee rate 1 Avg blocks to confirm 1.67 NumBlocks:% confirmed 1: 0.569 2: 0.844 3: 0.943 Fee rate 2 Avg blocks to confirm 1.33 NumBlocks:% confirmed 1: 0.715 2: 0.963 3: 0.989 Fee rate 3 Avg blocks to confirm 1.27 NumBlocks:% confirmed 1: 0.751 2: 0.974 3: 1.0 Fee rate 4 Avg blocks to confirm 1.25 NumBlocks:% confirmed 1: 0.792 2: 0.953 3: 0.994 Fee rate 6 Avg blocks to confirm 1.12 NumBlocks:% confirmed 1: 0.875 2: 1.0 3: 1.0 Fee rate 10 Avg blocks to confirm 1.09 NumBlocks:% confirmed 1: 0.901 2: 1.0 3: 1.0 Fee rate 30 Avg blocks to confirm 1.12 NumBlocks:% confirmed 1: 0.886 2: 0.989 3: 1.0 Alex
