confusing unit descriptions for "fee_per_mil" and "fee_rate"

[AndySchroder]

For fee_per_mil we have a definition of "The amount charged per milli-satoshis transferred expressed in millionths of a satoshi.". I interpret the units for this to be (1sat/1milli-sat)*(10^6 micro-sat)/(1sat)=(10^6 micro-sat)/(1milli-sat) .

For fee_rate we have a definition of "The effective fee rate in milli-satoshis. Computed by dividing the fee_per_mil value by 1 million.". If I take the above assumption, then I get ((10^6 micro sat)/(1milli-sat))/(10^6 microsat/1sat)=1sat/1milli-sat , which is not in milli-sat.

Should the definition of fee_per_mil read "The milli-satoshi amount charged per milli-satoshis transferred expressed in micro-satoshi per milli-satoshi"`?

Should the definition of fee_rate be "The milli-satoshi amount charged per milli-satoshis transferred, which is equal to fee_per_mil*10^6"? If this is correct, fee_rate is unitless.

Is there a clearer way to present it?

[AndySchroder]

Note, the above definitions came from ChannelFeeReport.

Note: lnd -h indicates:

bitcoin.feerate, The fee rate used when forwarding payments on our channels. The total fee charged is basefee + (amount * feerate / 1000000), where amount is the forwarded amount. (default: 1)

So, I'm even more confused at this point. If I look at a channel, using lncli feereport, fee_rate=0.000001 and fee_per_mil=1, which contradicts.

Also note, lncli help updatechanpolicy says "fee_rate, the fee rate that will be charged proportionally based on the value of each forwarded HTLC, the lowest possible rate is 0 with a granularity of 0.000001 (millionths)." This seems to be in agreement with ChannelFeeReport, but not lnd -h.

[Roasbeef]

Related to #1523

[yyforyongyu]

I was looking into this issue.

A bit confused though.

My understanding of the feerate comes from this line in link.go and this test. To put it here, the expected fee is calculated using,

f.BaseFee + (htlcAmt*f.FeeRate)/1000000

Since both f.BaseFee and htlcAmt are in msat, reversing the equation, the f.FeeRate is then using a uint as a billionth of a satoshi per msat, which is 'nano-sat' per msat, or, nsat / msat, this is more clear in the equation,

msat + (msat *  nsat / msat) / 1e6

And my understanding of the fee_rate is, the amount to charge in nsat to route 1 msat. If that's correct, I'd suggest two changes.

• The first is to fix the inaccuracies in the documentations, which is what this issue is about.
• A further approach, is to introduce a new type, lnwire.NanoSatoshi, to address the speciality of the fee rate used in routing payments. Also we could refactor some of the variable names.

[jhoenicke]

I think stating it in nanosat per millisat confuses people even more. It's a fraction. Just say its in "per million" (and don't abbreviate it as per_mil, since that may be read as per mille, which is just per thousand). We are used to compute in percent and this should follow this notion only it is in "per million" not "per hundred". There is also ppm (parts per million) but I'm not sure if it should be used here, since there are no particles.