[cryptography] bitcoin scalability to high transaction rates

Ian G iang at iang.org
Wed Jul 20 09:24:10 EDT 2011

On 20/07/11 8:02 AM, Sampo Syreeni wrote:
> On 2011-07-20, Ian G wrote:
>> To answer OP, typically all trading is done on a delayed and netted
>> settlement. Which is to say the trade might be done real time but the
>> settlement is batched for later, typically after market closing. No
>> money changes hands until later. This is especially true as you get
>> closer to liquidity or speculative trading, because of the nature of
>> the parties having no skin in the game, and trading on credit.
> Yes, and I should have touched this as well. The problem is, the biggest
> thing driving Bitcoin adoption is that it cuts down on the middlemen,
> and their cost. Government involvement, of course, but also the
> financial establishment. If you then have to rely on trusted third
> parties/marketplaces/settlement agency, even with Bitcoin, that
> nullifies half of the promise the currency has in the first place.

Yes, sure, but:

1.  we are talking about high frequency trading here, and speed is the 
first, second and third rule.  Each trade could be making 10k++ and up, 
which buys you a lot of leaches.

Basically, you have to get the trade down to the cost of a packet, delay 
and two secret key ops.  Indeed, if you can measure the delay of the 
secret key op, we might be encouraged to pre-calculate shared PRNG 
streams so as to speed up the encrypt/decrypt cycle.  (Gee I wonder if I 
should file a patent on that idea :P )

This and other aspects of high frequency trading forces a credit 
exposure to the trades, which requires someone to step in and control 
that credit.  No payments allowed to interfere with the trade itself. 
In the financial cryptography exchange for real time trading that I 
built, payments were delivered up-front, but there was intra-order 
trading that was done on credit by the central exchange.  That is, a 
sell order of 100 could be fulfilled 10 at a time until closed out. 
This was necessary to improve the liquidity, and as liquidity makes the 
trade happen or not in many cases, it dominated the question of credit 
and associated leaching costs.

If we're talking anything else like retail payments, then there is 
leeway to insist on pure BitCoin settlement at its speed.

2.  The payoff to the stationary bandit also closes the loop on the 
criminals you peer with.  In BitCoin, there is no such closure, you 
don't get to select your criminal partners.  This raises your costs.

> I also think the potential problem could be rectified rather simply:
> just build a suitable hash tree of transactions and only subject the
> root to the proof-of-work timestamping machinery, while offloading the
> millisecond by millisecond processing and storage to auxiliary sites.
> That way even high transaction densities would end up being a bona fide
> part of the shared log, but only summaries/hashes would need to be
> broadcast. Most of the hard, costly work of hashing and publicly storing
> those more frequent transactions could be done in a decentralized and
> less-trusted fashion, so that the middleman would be at least subjected
> to full competition.
>> I suppose we might try a bit-commit style of bilateral exchange but it
>> would need to overcome the speed and cost advantages of the TTP.
> The Bitcoin economy seems to work somewhere in between. Both in
> efficiency, and as everybody knows, privacy as well. At least to me the
> question then is, is Bitcoin really at the Pareto frontier with regard
> to efficiency, privacy, latency and whathaveyou, at the same time. I'm
> not too sure it is.

Well, it's clearly inefficient, but that's a design feature :)  Privacy 
can't really be claimed as it has a public database, and it's a sucker 
for datamining.  Latency I gather has its issues too.

If one were to speculate as to some sort of frontier of benefits, then 
I'd say when it came to trading, BitCoin would be a distance 10th.  A 
psuedonymous system is far more efficient and secure, and adding 
chaumian blinding adds a modicum of untraceability [0].

No centralised control of the issue is a benefit of BitCoin, but we 
would need multiple currencies, grounded in contracts.  (Distributed 
issuers in open issuance of contracts achieves 98% of the distributed 
bounty of BitCoin.)

>> It's not likely that the remaining of the population could appreciated
>> it, sure :)
> ...  And then, if
> you can't spend willy-nilly, anything you just keep off the market is as
> good in the medium run as something that never existed in the first
> place; money really doesn't help you much unless you can spend it, which
> for the most part makes the entitlement worries moot.

Exactly.  This is where hoarding meets Jon's Highlander constant meets 
Fort Knox.


[0] Chaumian blinding is only untraceable on paper.  In practice, the 
untraceability depends on many implementation and economics factors.

More information about the cryptography mailing list