search

thrumdev / nomt

Apache License 2.0
96 stars 5 forks source link

opt: pre-sort writes #486

Closed rphmeier closed 2 weeks ago

rphmeier commented 2 weeks ago

This optimization is a bit more tentative.

The overhead of sorting is relatively low, but it seems to help with the throughput of HT writes in particular.

rphmeier commented 2 weeks ago

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @rphmeier and the rest of your teammates on Graphite Graphite

deuszx commented 2 weeks ago

Can you share your testing methodology for this? IE which tests in particular you run to uncover these improvement options? I'll appreciate it a lot.

rphmeier commented 2 weeks ago

Sure. In this case I used benchtop (located in the benchtop folder) with a couple of extra printlns added in key points to uncover IOPS (e.g. bitbox::writeout and beatree::writeout)

I generated a database with 2^25 items and ran some test cases against it.

pepyakin commented 2 weeks ago

Merge activity