* Re: [bitcoin-dev] Bitcoin Guarantees Strong, not Eventual, Consistency.
2016-03-02 8:39 ` Natanael
@ 2016-03-02 16:56 ` Emin Gün Sirer
0 siblings, 0 replies; 3+ messages in thread
From: Emin Gün Sirer @ 2016-03-02 16:56 UTC (permalink / raw)
To: Bitcoin Dev
[-- Attachment #1: Type: text/plain, Size: 3604 bytes --]
> The entire point of the definition of eventually consistency is that your
> computer system is running continously and DO NOT have a final state, and
> therefore you must be able to describe the behavior when your system either
> may give responses to queries across time that are either perfectly
> consistent *or not* perfectly consistent.
>
This is not the definition of eventual consistency. From
https://en.wikipedia.org/wiki/Eventual_consistency:
Eventual consistency is a consistency model used in distributed computing
to achieve high availability that informally guarantees that, if no new
updates are made to a given data item, eventually all accesses to that item
will return the last updated value.
The actual definition makes it quite clear that a system need not have a
final state to be evaluated for its consistency properties. Almost all
practical database systems execute continuously without a final state.
> And Bitcoin by default *does not* ignore the contents of the last X
> blocks. A Bitcoin node being queried about the current blockchain state
> WILL give inconsistent answers when there's block rearrangements = no
> strong consistency.
One could split hairs here by pedantically defining "Bitcoin by default" --
you could refer to just the reference client code and ignore the shim code
in the app that interfaces with the client -- but that'd drag us into a
fruitless email-list-style discussion from which no one would emerge any
wiser. I'll avoid that, and will instead dryly note that the reference
client's listreceivedbyaddress will return the number of confirmations by
default, and every application will then check the confirmations value to
confirm that it exceeds that application's own omega, while
getbalance,getreceivedbyaddress will take a number of confirmations as an
argument, shielding the app from reorgs of the suffix. That is precisely
the point made in the post.
> Not to mention that your definition ignores the nonzero probability of a
> block rearrangement extending beyond your constant omega.
>
The post covers this case. Technically, there is a difference between 0
probability and epsilon probability -- this is the reason why Nakamoto
Consensus was an exciting breakthrough result; the same reason why
Lamport's results regarding a 3f+1 bound on the Byzantine Generals Problem
do not apply to Nakamoto Consensus; and the same reason it took our paper
(Majority is Not Enough) to show that Nakamoto consensus has a similar 33%
bound as Lamport-style consensus when it comes to tolerating Byzantine
actors.
Practically, however, there is little difference between 0 and a value that
exponentially approximates 0, given that we operate on hardware subject to
random errors. The post makes the case that one can pick an omega such that
the probability of your processor mis-executing your code is larger than
the probability of observing a reorganization.
Bitcoin provides a probabilistic, accumulative probability. Not a perfect
> one.
>
Sometimes, non-technical people get confused about the difference between
very very very small probabilities that approximate 0 and 0. For instance,
some people get very worried about hash collisions, on which Bitcoin relies
for its correctness, whose probability also drops exponentially but is not
exactly 0. Your overall point seems to be an analogous concern that
Bitcoin's exponentially dropping probability of reorganization isn't quite
a "perfect" 0. If so, I agree and the original post made this quite clear.
Though I hope we can avoid that kind of discussion on this particular list.
- egs
[-- Attachment #2: Type: text/html, Size: 4868 bytes --]
^ permalink raw reply [flat|nested] 3+ messages in thread