Skip to content

Improve rolling fast filter

Peter Tschipper requested to merge ptschip/nexa:dev_fastfilter into dev

The current rolling fast filter has a high false negative rate and causes the repeated resending of inv's to nodes that we just received inv's from; By using a generational approach we can avoid this issue caused by erasing data from the filter and so we can use this rolling filter in place of a regular bloom filter and improve overall performance under load significantly.

Still needs a bit of work adding benchmark tests and cleaning up. Probably I should leave the old rolling filter in place so we can see the performance difference in the benchmark tests.

Merge request reports