Relay network design - Githubissues

MatrixAI / Relay

Service-Centric Networking for Matrix Automatons

0 stars 0 forks source link

Relay network design #4

Open ramwan opened 6 years ago

ramwan commented 6 years ago

Things to examine when determining how a relay network will be created and how communication will occur on top of the network are:

p2p networks and protocols
modularity of network stacks and infrastructure
algorithms and protocols

This issue builds off a decent amount of what Issue: Relay Service Abstraction and Addressing attempts to deal with.

P2P networks and protocols

dealing with boxes behind NATs
distributed services
- ephemeral/persistent services
- ephemeral/persistent automatons

Network modularity

reducing duplication and redundancy
allowing "plug and play" functionality for changing specs and requirements

Protocols and algorithms

randomisation
ability to incorporate arbitrarily created protocols and translate specs into networking implementations

CMCDragonkai commented 6 years ago

@Ramwan please write down the issues we discussed with regards to Architect design and Automaton specification and the composition of Automatons, and the requirements to be able to move Automatons around nodes without interrupting communication up to the constraint what is an atomic message. Let's enumerate all the kinds of "zero downtime changes" that can occur for composed Automatons. I can think of some examples already: an automaton fails by breaking, composition objective has changed, scale up, scale down, optimisation has decided that an automaton should be moved somewhere else. There's more!

ramwan commented 6 years ago

Another thing to consider is how services can be migrated and state managed during and after the migration. We'll need to consider how this is done and what trade-offs we'll take. Tracked in issue #12.

ramwan commented 6 years ago

Initial design considerations:

after researching the architecture of Serval and Mobile IP, I've found that the abstractions in both are very similar and simple enough concept wise to work well and hopefully be implemented. As such I've started writing a design doc to flesh out my thoughts and to have them examined
To test this design, I will be trying to implement the simplest case of migrating a stateless container responding to pings. This will give us a clearer picture of exactly how migration could be handled and any design flaws may begin to pop up.

ramwan commented 6 years ago

writeup for initial experiment @ https://github.com/MatrixAI/Relay/blob/master/experiments/stateless_container_ping_migrations/README.txt