DistOS 2021F 2021-11-30

From Soma-notes
Jump to navigation Jump to search


Lecture 20

We've discussed blockchains & bitcoin
What's different with Ethereum?
 - "smart contracts"
 - distributed application running in VMs
    - with a distributed, consistent data store

Key difference - who runs the code is determined by economics
 - who is willing to pay

The infrastructure for running smart contracts doesn't have a lot
of resources left over for the actual contracts
 - most goes to proof of work

So we have something like WebAssembly for running code on the nodes
 - and this code can then interact with other Ethereum accounts,
   and potentially other services that have interfaces into Ethereum

What does proof of work really buy you?
 - compared to other systems we've discussed?

With blockchain technologies, we make a bargain:
 - give up any notion of performance or efficiency
    - proof of work consumes huge amounts of resources
 - in exchange, no trusted third party
    - but what do you have instead?

Instead, trust majority of computing power in the network + correctness/security of code, algorithms, & protocols
 - why is this better than a third party?

What do you trust more, law/social arrangements or code?
 - but this really isn't the choice
 - because the code *will* have flaws, and then they'll be
   addressed using laws & social arrangements

What is a fork?
 - disagreement over what the "valid" chain is
    - supposed to be the one with the most compute on it
    - but really, could be any chain as long as everyone agrees

Realize the blockchain is just a consensus log
 - canonical order of events (remember Delos)

Partitions are when consensus breaks down
 - that's all a fork is

Note that with NFTs the things being traded aren't in the blockchain
 - just the hash

Secure hashes have a few key properties
 - If I have a hash, I can't create data that has that hash
    (can't reverse it)
 - Hashes of two sets of data won't be equal
    - so different hashes => different data
    (no collisions)

So, a hash can serve as an unforgeable identifier for digital data

That's why when you download programs they often post hashes of the files
  - you can check to make sure your downloaded file has the same hash as posted on the website
  - digital signatures are all based on secure hashes

If I "own" the hash of some data, that's equivalent to proving I
"own" the data
 - but we can have a weird situation where I can prove ownership
   of data nobody has ever seen

When you get rid of trusted third parties, nobody can step in when bad things happen

I think there is a future in blockchain technologies
 - but it will be proof of stake or, more likely,
   "permissioned" blockchains (i.e., with trusted third parties)

Trust is really about what happens when things break
 - who is responsible
 - who will fix it, or be punished for its failure

We aren't good at thinking about failure, particularly in an adversarial environment

What happens if someone finds a way to make collisions in SHA3 efficiently?

Is this unlikely?
 - in my lifetime, MD5 & SHA1 went from being secure to
   people finding collisions

Cryptographic primitives have NO PROOF of security
 - block ciphers
 - secure hashes
 - public key cryptography
 - MACs

Financial systems only work when people believe in them
 - the moment that faith is lost the system breaks