search

celestiaorg / ipld-plugin-experiments

MIT License
5 stars 4 forks source link

Switch to LL-specific plugin or run LL node #10

Closed liamsi closed 2 years ago

liamsi commented 3 years ago

Use this plugin when tackling #4 and #5.

An alternative to explicitly use the plugin, could be to actually spin up a lazyledger-app node and run the experiments directly on the embedded IPFS node using the API that will be introduced as an outcome of lazyledger/lazyledger-core#170. This approach could also be used via CI, e.g. triggered on each PR merged to master (in lazyledger-app/-core) or weekly, or, before every release. In the beginning, monthly and some way to manually trigger the CI run should be sufficient.

Copy & pasting a convo from discord related to figuring out why numbers differ with the MVP (ref: https://github.com/lazyledger/lazyledger-core/issues/377). Quoting @musalbas:

Here's a potential process to check.

  1. Switch the experiments to the NMT plugin. Does it still work for 1x32? If not, the issue is with the plugin.
  2. Make the experiment do a 32x32 square instead of 1x32. Does it still work? If not, the issue is indeed with ipfs and we should figure out what changes because we're publishing more data roots at the same time.

Adding this here as there is some overlap with 1. and this issue: LL nodes use the NMT plugin but not with a single data root.

Switching the experiments to just use a NMT plugin would need to happen first. Unless we keep this repo as simple as it currently is use something else to actually spin up LL nodes and do measurements.

liamsi commented 3 years ago

I would also add that in above's potential process, there is something that also needs to be considered:

Another dimension that isn't factored into this is that in the MVPs case: blocks were produced rapidly. Originally, even as fast as the machine could, usually < 1 sec. That is just loads of data from the pov of the dht. After adding the dummy-app there is an artificial delay to simulate execution and reaching consensus but it is also just a second. Note, this delay can be adjusted via a command line param. (ref: https://github.com/lazyledger/lazyledger-core/issues/383) The experiments in December/January where like one round of block production only. Which I think matters as well.

To actually come to a conclusion if this has anything to do with the NMT plugin, then this should also be factored in.

liamsi commented 2 years ago

I don't think we need to continue these experiments but should instead use our node software to experiment with different sizes etc. I'm closing this for now (cc @Bidon15 @jbowen93).