apfitzge / prio-graph Goto Github PK

Add benchmarks for "tree-style" access:

• configurable number of trees
• configurable (randomized) number of branches
• configurable (randomized) branch length
• varying number of accounts per transaction

There have been breaking changes since the previous release.
This issue is a reminder that the version must be bumped in the next release to 0.3.0.

Update random access benchmarks to use a specfic number of accounts per transaction.

• Resizing of HashSet while iterating is very expensive

Migrate the existing benchmarks to use criterion. This provides more detailed reports on specific benches and just makes it generally easier to notice regressions or improvements to performance.

Add benchmarks for

• all non-conflicting transactions
• with variable numbers of accounts per transaction

• block_tx! was originally a closure
• compile was not convinced that mutability was respected
• converted it to a macro...which is ugly.

Try to find a way to remove the macro without runtime costs via RefCell or similar

Want to more reliably test preformance related changes, and the current benchmarks are not sufficient.

• #60
• #65
• #61
• #62
• #63

Add benchmarks for:

• All transactions conflict with each other
• With varying numbers of accounts

• If there are multiple reads that conflict with a write, only the first one is counted as conflicting.