Message-ID: <52E032BD.4020206@bcdev.net>
Date: Wed, 22 Jan 2014 22:06:05 +0100
From: bc <bc@bcdev.net>
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
	rv:24.0) Gecko/20100101 Thunderbird/24.1.0
MIME-Version: 1.0
To: bitcoin-development@lists.sourceforge.net
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Subject: [Bitcoin-development] Combining big transactions with hash-only
	blocks to improve tps.
Precedence: list

Pdf version:
http://bcdev.net/data/bitcoin_big_tx_with_coin_join.pdf


== Combining big transactions with hash-only blocks to improve tps. ==

==== Abstract: ====
I've heard people talk about including only hashes in a block to speed 
up the network and also about using CoinJoin to improve privacy. I've 
not heard anyone talk about implications of combining these two 
techniques. I think that it would both improve network's anonymity, but 
also improve tps by a few orders of magnitude.

I propose two optimizations:
1. Keep only hashes of transactions included in a block. Transfer all tx 
separately.
2. Use CoinJoin to merge transactions from many users for online 
shopping and banking.
3. Use Jumbo transactions as a fallback for applications where CoinJoin 
is inappropriate.

==== Keeping only hashes of tx in a block: ====
Currently every bitcoin block includes a copy of all transactions. This 
is redundant and unnecessary, since after the transaction gets 
transmitted, every node learns about it in seconds.
By keeping only transaction hashes in block, we can keep block 
propagation time from increasing.
Assuming a typical tx with one or two inputs and two outputs [typically 
300 bytes], current 1MiB block can contain about [assuming a block every 
10 minutes]:
1MiB / 300 bytes = 3300tx = 5.5tps

By keeping only hashes in a block [32 bytes per hash]:
1MiB / 32 bytes = 31000tx = 50tps

== Benefits: ==
This method allows to achieve more tps without increasing the block 
propagation time, which is critical for mining decentralization.
It removes redundancy, since every tx has to be transmitted only once.
It leads to a more consistent bandwidth utilization [large transactions 
are transmitted all the time, while blocks are kept small and easy to 
propagate].
Because a block size is a constant, mining fees would not depend on the 
size of a transaction. Obviously to limit the network flood, there 
should be a transaction size limit.

== Problems: ==
Selfish miner can keep a subset of transactions only for yourself and 
release them only with a new block. This problem can be mitigated by 
making nodes verify all transactions before propagating a block. The 
incentive will then be to mine only a well-distributed transactions to 
lower orphan rate.
The miner can try to sneak up invalid transaction in a block. This 
problem is also mitigated by not accepting a block before it gets verified.

==== CoinJoin: ====
If the block size keeps only hashes, a transaction can be much bigger. 
Since CoinJoin allows many people to send coins with one transaction, 
the effective transaction rate can be increased considerably.

== Example: ==
Let's assume the transaction size limit of 50KiB. Limit of this size 
allows for a CoinJoin transaction between 50KiB / 300b = 170 participants.
So for a block of 1MiB, it would allow for 50tps * 
170effective_transactions/tx = 8500tps.

== Benefits: ==
There would be an incentive for users to use CoinJoin by default [lower 
tx fees per effective transaction], which would greatly increase 
anonymity of the network.
Since block size stays the same, block propagation time also stays the same.
It doesn't require any changes to the protocol. CoinJoin transactions 
were always supported in bitcoin.

== Problems: ==
1) CoinJoin requires collaboration between many users in real-time. It 
means, that transaction must be distributed to every CoinJoin 
participant, and every participant has to sign it before it can be 
released. Therefore it induces delays, which can take some time.
It wouldn't be an issue with Internet banking or on-line shopping [where 
even 10 minutes per transaction is fast enough], however even 20 seconds 
can make the system unsuitable for POS payments.
Potential solution: Use bigger CoinJoin user base for online payments 
[with smaller fees], and a smaller one for POS payments [with larger fees].

2) Signing a CoinJoin transaction requires to transfer a whole 
transaction for a user to sign.
This can sometimes take up to a few minutes on a very slow networks.

3) CoinJoin transactions are limited. They are good enough for money 
transfer, but for more advanced appliances CoinJoin might be inadequate.

==== Jumbo transactons: ====
I propose another tx type as a fallback where CoinJoin is not Combining 
big transactions with hash-only blocks to improve tps.applicable. It 
would remove the CoinJoin induced delays, while keeping transaction 
sizes big.

Image: http://bcdev.net/data/jubo_transaction_description.png

Transaction joiner is a service that collects transactions from clients 
and publishes them as a Jumbo transaction.
Jumbo pubkey prevents transaction from being modified. It can only be 
accepted or rejected by the miner as a whole, which should limit 
discrimination.

== Algorithm: ==
1) Transaction joiner sends a Jumbo pubkey hash to the client.
2) Client creates a transaction, includes a Jumbo pubkey hash and signs it.
3) Transaction joiner waits until there are enough transactions and 
releases a Jumbo transaction to the network.
4) A miner includes only a hash of a Jumbo transaction in a block, he 
cannot cherry-pick individual transactions from the bulk.
5) The network checks if every transaction inside a Jumbo transaction 
includes a Jumbo pubkey hash and if every transaction inside is valid.

== Benefits: ==
Since the block size stays the same, block propagation time also stay 
the same.
There is no need to wait for every participant to sign the transaction. 
It's therefore more suitable for POS payments.
No additional network overhead for a thin client compared to a standard tx.
Backwards compatibility with current transaction system.

== Problems: ==
1) Jumbo transactions don't mix coins. Anonymity of the network is not 
increased.
2) There would be an incentive to use this transaction type by default 
[compared to CoinJoin].

Potential solution:
Make Jumbo transaction size limit lower than CoinJoin. That would make 
fees for these transactions higher, thus creating an incentive to only 
use them when necessary.

3) Transaction joiner has to wait for a Jumbo transaction to be big 
enough before it gets released.
It's not a big problem. When the network load is low, the fee required 
for a tx to be included should be lower, allowing for smaller Jumbo 
transactions. When the network load is high, it takes less time to fill 
a Jumbo transaction.

==== References: ====
Increasing the Network Hashing Power by reducing block propagation time 
https://bitcointalk.org/index.php?topic=145066.0

CoinJoin: Bitcoin privacy for the real world
https://bitcointalk.org/index.php?topic=279249.0

Bitcoin: A Peer-to-Peer Electronic Cash System
http://bitcoin.org/bitcoin.pdf