Skip to content

feat: multi-provider fan-in #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
b5 opened this issue Oct 24, 2024 · 4 comments
Open

feat: multi-provider fan-in #4

b5 opened this issue Oct 24, 2024 · 4 comments
Assignees
Milestone

Comments

@b5
Copy link
Member

b5 commented Oct 24, 2024

download_with_opts currently allows multiple node addresses, but fetches in series. download_with_opts should be modified/expanded to support dividing the request up across multiple nodes, fanning the goodput responses in to a single response stream.

@arilotter is currently adding this on her own downstream out of urgent need. We should try to help, at least through code guidance, considering we've wanted this in iroh blobs for some time.

@rklaehn
Copy link
Collaborator

rklaehn commented Nov 8, 2024

So you just want to do a fixed plan, not some dynamic thing?

@arilotter
Copy link
Contributor

arilotter commented Nov 11, 2024

So you just want to do a fixed plan, not some dynamic thing?

being able to add a second node as a download source after the blob starts downloading from a first node would be really useful - a dynamic thing seems like it would be awesome long-term, but something fixed like "if you have 2 nodes, download half the file from each" would be way better than nothing

@n0bot n0bot bot added this to iroh Jan 23, 2025
@rklaehn rklaehn self-assigned this Feb 4, 2025
@rklaehn rklaehn moved this to 🏗 In progress in iroh Feb 4, 2025
@rklaehn
Copy link
Collaborator

rklaehn commented Feb 4, 2025

Milestone 1: assume providers have all the data

  • figure out open ended intervals
  • write an actual example instead of just a test
  • make the state machine more solid

Bonus:

  • write a good, simple scheduler
    • prioritize single source chunks? Multiple downloads per peer?
  • write a store bitfield watcher

Milestone 2: providers don’t have all the data

  • extend protocol to allow remote bitmap watching

Milestone 3: solidify

  • make sure the driver does not do unnecessary connect calls
  • make the driver more solid
  • use tokio JoinSet instead of task maps?

@amiyatulu
Copy link

Also, prioritize the nodes based on their bandwidth.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Status: 🏗 In progress
Development

No branches or pull requests

5 participants