Revisit fee structure

[tromp]

The current minimum relay fee rules use

accept_fee_base = 1000000

as set in the default grin-server.toml.
Since this is in units of nanogrin, it expresses a base cost of 1 milli-grin.
This is what you pay for each unit of a transaction weight, which is defined in

grin/core/src/core/transaction.rs

Lines 847 to 856 in 20e5c19

	/// Calculate transaction weight from transaction details. This is non
	/// consensus critical and compared to block weight, incentivizes spending
	/// more outputs (to lower the fee).
	pub fn weight(input_len: usize, output_len: usize, kernel_len: usize) -> usize {
	let body_weight = output_len
	.saturating_mul(4)
	.saturating_add(kernel_len)
	.saturating_sub(input_len);
	max(body_weight, 1)
	}

as the linear combination 4 * #outputs + 1 * #kernels + (-1) * #inputs, with a minimum of 1.
This means that a typical tx with 2 outputs, 1 input and 1 kernel requires a relay fee of 8 milligrin,
currently worth about 1/3 cent.

We should revisit this base fee and tx weight coefficients to provide a good long term balance
between spam resistance and tx affordability. It would also be nice to make the tx weight coefficients seem less arbitrary.

[lehnberg]

Is there some fee structure that could promote privacy preserving output selection? Recall the discussion in #2186 for instance.

For example, what if we could inceltivize 'balanced' transactions? (two inputs spend to two outputs), or some other structure that makes the tx graph better obfuscated.

[tromp]

Yes, we should consider incentivizing privacy enhancing transactions.
A tx with 2 inputs and 2 outputs is already cheaper in fees than a tx with only 1 input and 2 outputs.

[lehnberg]

#3481

[antiochp]

Closing - included as part of 5.0.0.