Resources

• Analysis of RSpace data layer

• Related conversation in #RSpace

• Related conversation in #RSpace

• 

Discussion about Dom’s findings

• Stop storing sequences. Start storing pieces of sequences (ex hashes)

• If we don’t address the problem of list growth now, we will have to at some point.

• 
  

What could we do?

• Issue in the storage format. Recommendation to fix the AST

  • Serialization

  • How RSpace is built to…

• First address how we represent the things we store

  • Should be provided on the Rholang side

    • Incoder and decoder

• Optimize waiting continuations

Discussion about garbage collection

• Renovated RSpace can manage the hash change

• What’s the use case for garbage collection?

  • To support new nodes catching up

• What is garbage?

  • Data that is old (ex on the other side of finalization) and/or inaccessible

  • DECISION run garbage collection + defined grace period

• Three types of garbage to consider

  • Orphaned blocks

  • Unmatched names transferred from hot store to history trie

  • Names that are no longer referenceable after finalization

    • Requires tooling to retrospect Rholang and ask Casper about finalization

Next steps

• Fix the example contracts so they don’t use single channels (ex ListMap)

• To support further investigation, measure the size of data being moved between layers

• Implement a binary encoder and decoder off of the binary scodec

• Discuss if Casper can share information about orphaned

• Define heuristic for when to collect garbage

  • Define a heuristic for finding out what’s unreferencable: reverse mark and sweep