Owners: @Mehmet @Daniel Sanchez Quiros
Support: @Álvaro Castro-Castilla
Conclusions
- NomosDA is a better solution for us.
- The reason why NomosDA (with FK20 of course) is better for us isn't trivial, but in short:
- Having many columns is good for decentralization. NomosDA clearly performs better with many columns. We knew this, but until this last version with corrected numbers, things didn't seem practical.
- A huge point is that NomosDA has a substantially lower ANBC. This is also key for decentralization. This is even more important than the previous point. The reason is that it will unlock the possibility of having thousands of Zones.
- NomosDA scales gracefully and predictably from 32kb to 1024kb, which means that we can have very small Zones and very large Zones using the same protocol.
- NomosDA offers the possibility of having lightweight zones, using a single thread, which also opens up an interesting design space.
Rationale for analyzing and selecting the DA scheme
These are the main criteria:
- Maximum decentralization: low RB + low ANBC with 32 kb blob. This allows the creation of many zones.
- High-enough throughput: there should be a mechanism to support at least 100kb/s in a high-performance zone. Ideally we would reach 250kb/s.
- Operate with nodes behind NAT networks, in a say similar to a P2P software like BitTorrent.
- Latency: while we are comparing mainly based on bandwidth terms, latency is detrimental to UX, so it should be also taken into account.
- Flexibility: supporting both very large and high-performance Zones along with very small and low-performance
Generally, this is what we would expect as the number of connections that a basic home setup would offer:
- Desktop: 500 - 1,000 connections
- Router: 1,000 - 2,000 connections
This would be a mid-range setup: