On Sat, May 16, 2015 at 1:22 AM, Rusty Russell <rusty@rustcorp.com.au> wrote:

Some tweaks:

1) Nomenclature: call tx_size "tx_cost" and real_size "tx_bytes"?

Fair enough.

2) If we have a reasonable hard *byte* limit, I don't think that we need
the MAX(). In fact, it's probably OK to go negative.

I agree, we want people to compress the UTXO space and a transaction with 100 inputs and one output is great.

It may have privacy problem though.

3) ... or maybe not, if any consumed UTXO was generated before the soft
fork (reducing Tier's perverse incentive).

The incentive problem can be fixed by excluding UTXOs from blocks before a certain count.

UTXOs in blocks before 375000 don't count.

4) How do we measure UTXO size? There are some constant-ish things in
there (eg. txid as key, height, outnum, amount). Maybe just add 32
to scriptlen?

They can be stored as a fixed digest. That can be any size, depending on security requirements.

Gmaxwell's cost proposal is 3-4 bytes per UTXO change. It isn't 4*UXTO.size - 3*UTXO.size

It is only a small nudge. With only 10% of the block space to play with it can't be massive.

This requires that transactions include scriptPubKey information when broadcasting them.

5) Add a CHECKSIG cost. Naively, since we allow 20,000 CHECKSIGs and
1MB blocks, that implies a cost of 50 bytes per CHECKSIG (but counted
correctly, unlike now).

This last one implies that the initial cost limit would be 2M, but in
practice probably somewhere in the middle.

tx_cost = 50*num-CHECKSIG
+ tx_bytes
+ 4*utxo_created_size
- 3*utxo_consumed_size

> A 250 byte transaction with 2 inputs and 2 outputs would have an adjusted
> size of 252 bytes.

Now cost == 352.

That is to large a cost for a 10% block change. It could be included in the block size hard fork though. I think have one combined "cost" for transactions is good. It means much fewer spread out transaction checks. The code for the cost formula would be in one place.