For the Network Architecture ART report I’m collecting questions, topics, concepts to define, etc.
Reply to this post with if you have anything in mind that should be covered in the report.
1 Like
Feedback in general:
- I think the overall architecture as outlined here roughly makes sense.
- I’d really like to see clearer definitions and mathematical semantics for the abstractions in play here. This report (currently) uses a lot of concepts which are not precisely defined and I think it’s often not going to be clear to a lay reader what is meant.
- I think diagrams of how topics + data structures + storage + pub/sub etc. work within the domains would be really helpful for conveying the design intentions here.
Specific questions and comments:
Asynchronous user-to-user communication is made possible by encrypted inboxes hosted by user-designated nodes.
As in, store-and-forward, with some storage period defined by whatever the user has agreed upon with their designated node?
A node consists of a set of running engine instances, has a cryptographic identity and a number of transport addresses. Nodes communicate with each other over authenticated and encrypted transport channels.
Nodes could in principle have multiple cryptographic identities, right? (I suppose that from the external perspective these would be “different nodes”, but we should clarify the definition here).
A distributed data storage system allows storing immutable and mutable data in the network. Immutable data is encrypted and content-addressed, while mutable data uses encrypted pub/sub channels that store and forward an append-only log of messages that form a causal DAG backed by a Blocklace [AlSh24] data structure. Pub/sub messages may contain partially ordered application messages, references to immutable data, or operations on Mergeable Replicated Data Types (MRDTs) [KPSJ19] and Conflict-free Replicated Data Types (CRDTs) [SPBZ11].
What exactly do you mean by “mutable” data? An “append-only log of messages” doesn’t sound “mutable” to me. Do you mean that we interpret that append-only log as a sequence of updates to a single structure (which could then be considered “mutable”)?
type UserAdvert :=
mkNodeAdvert {
id : UserID;
nodes : Set NodeAdvert;
relays : Set NodeAdvert;
prekeys : Set PreKey;
version : Nat;
created : AbsTime;
sig : Commitment;
};
The type and names in the data structure seem to mismatch (should it be mkUserAdvert)?
In a domain, the set of available mutable data structures are published in an index, which contains a map of referencable data paths to cryptographic topic IDs that contain the data structure.
What is a “referencable data path”?