celestia-node
celestia-node copied to clipboard
share: benchmark `Read/WriteEDS`
Introduced #1104 and #1105 are allocations and CPU-intensive serialization functions. We should know how they perform and always have a starting point on where to improve further.
A draft of optimizations tbd for WriteEDS:
- [ ] Use sha256-simd
- [ ] Use buffer pool for CAR allocations
- [ ] Avoid copying the EDS
- Write shares/samples at the moment of iterating