Open jpountz opened 5 months ago
For now we can use ramhog for this if needed. In the future we could specify cgroupv2 limits in the systemd service/timer for all the benchmarks, or run particular benchmarks with systemd-run
, specifying lower limits.
@rmuir also suggested using fincore --output-all /path/to/index/*
to monitor how many hot/cold pages we see in the index while/after benchmarking.
We're currently working on better parallelizing I/O for Lucene queries. To be able to track improvements (and regressions!), it would be nice if we could run nightly benchmarks with a page cache size that is only a fraction of the total index size.