Status
- 2022-11-18: This draft is ready for wider consumption and comment. IPFS Stewards expect to review it with @Juan Benet and finalize a plan as part of ‣.
- 2022-11-17: The outline of the doc has been started. A draft for more general consumption will be by EOD 2022-11-18. @Steve (biglep) will put a new status update when the draft is ready (and will advertise in a couple of Slack channels).
Context
PL infra costs are understandably being reviewed/scrutinized. One area that has come up are the hydra boosters that PL operates. For example, here was the 202210 cost:
Hydras have traditionally been seen as value across a couple of angles:
- Angle 1 - Improving network performance
- Angle 2 - Network bridging (to Indexers)
Questions have come in though around “can't we just shut them down since those angles may not be holding up as much now and since we have alternatives”. Specifically:
- They don't appear to help speedup the network much (per https://www.youtube.com/watch?v=zhzxJGoLTg0 )
- It is possible to add cid.contact as a default Reframe content router directly in Kubo.
Below is analysis, discussion, and plans on how we can reduce costs here.
Performance impact of dialing down hydras (ignoring bridging concerns)
Anticipated Impact
[Summary of results based on Hydras: Reliability and Effectiveness]
1. What we know today
Experiment Summary: we have published CIDs from nodes we controlled and pushed the provider records to the 20 closest peers to the CID. We then requested the CIDs from the DHT network and experimented with: