On Tue, May 26, 2015 at 12:03:09AM +0200, Mike Hearn wrote:
> CPFP also solves it just fine.

CPFP is a significantly more expensive way of paying fees than RBF,
particularly for the use-case of defragmenting outputs, with cost
savings ranging from 30% to 90%


Case 1: CPFP vs. RBF for increasing the fee on a single tx
----------------------------------------------------------

Creating an spending a P2PKH output uses 34 bytes of txout, and 148
bytes of txin, 182 bytes total.

Let's suppose I have a 1 BTC P2PKH output and I want to pay 0.1 BTC to
Alice. This results in a 1in/2out transaction t1 that's 226 bytes in size.
I forget to click on the "priority fee" option, so it goes out with the
minimum fee of 2.26uBTC. Whoops! I use CPFP to spend that output,
creating a new transaction t2 that's 192 bytes in size. I want to pay
1mBTC/KB for a fast confirmation, so I'm now paying 418uBTC of
transaction fees.

On the other hand, had I use RBF, my wallet would have simply
rebroadcast t1 with the change address decreased. The rules require you
to pay 2.26uBTC for the bandwidth consumed broadcasting it, plus the new
fee level, or 218uBTC of fees in total.

Cost savings: 48%


Case 2: Paying multiple recipients in succession
------------------------------------------------

Suppose that after I pay Alice, I also decide to pay Bob for his hard
work demonstrating cryptographic protocols. I need to create a new
transaction t2 spending t1's change address. Normally t2 would be
another 226 bytes in size, resulting in 226uBTC additional fees.

With RBF on the other hand I can simply double-spend t1 with a
transaction paying both Alice and Bob. This new transaction is 260 bytes
in size. I have to pay 2.6uBTC additional fees to pay for the bandwidth
consumed broadcasting it, resulting in an additional 36uBTC of fees.

Cost savings: 84%


Case 3: Paying multiple recipients from a 2-of-3 multisig wallet
----------------------------------------------------------------

The above situation gets even worse with multisig. t1 in the multisig
case is 367 bytes; t2 another 367 bytes, costing an additional 367uBTC
in fees. With RBF we rewrite t1 with an additional output, resulting in
a 399 byte transaction, with just 36uBTC in additional fees.

Cost savings: 90%


Case 4: Dust defragmentation
----------------------------

My wallet has a two transaction outputs that it wants to combine into
one for the purpose of UTXO defragmentation. It broadcasts transaction
t1 with two inputs and one output, size 340 bytes, paying zero fees.

Prior to the transaction confirming I find I need to spend those funds
for a priority transaction at the 1mBTC/KB fee level. This transaction,
t2a, has one input and two outputs, 226 bytes in size. However it needs
to pay fees for both transactions at once, resulting in a combined total
fee of 556uBTC. If this situation happens frequently, defragmenting
UTXOs is likely to cost more in additional fees than it saves.

With RBF I'd simply doublespend t1 with a 2-in-2-out transaction 374
bytes in size, paying 374uBTC. Even better, if one of the two inputs is
sufficiently large to cover my costs I can doublespend t1 with a
1-in-2-out tx just 226 bytes in size, paying 226uBTC.

Cost savings: 32% to 59%, or even infinite if defragmentation w/o RBF
              costs you more than you save

-- 
'peter'[:-1]@petertodd.org
0000000000000000134ce6577d4122094479f548b997baf84367eaf0c190bc9f