results under higher memory pressure show that tokio-epoll-uring pays off

Browse Source

setup:

sudo mkdir /sys/fs/cgroup/benchmark
admin@ip-172-31-13-23:[~/neon-main]: sudo mkdir /sys/fs/cgroup/benchmark
admin@ip-172-31-13-23:[~/neon-main]: sudo chown admin:admin /sys/fs/cgroup/benchmark
admin@ip-172-31-13-23:[~/neon-main]: sudo chown admin:admin /sys/fs/cgroup/benchmark/cgroup.procs
admin@ip-172-31-13-23:[~/neon-main]: echo THE_PID_OF_THE_SHELL_WHERE_WE_LAUNCH_PAGESERVER > /sys/fs/cgroup/benchmark/cgroup.procs

from another shell, that's not in the cgroup, run pagebench

admin@ip-172-31-13-23:[~/neon-main]: RUST_BACKTRACE=1 ./target/release/pagebench get-page-latest-lsn --mgmt-api-endpoint http://localhost:15011 --page-service-connstring=postgresql://localhost:15010  --keyspace-cache keyspace.cache  --per-target-rate-limit 2000 --limit-to-first-n-targets 500  --set-io-engine YOUR_IO_ENGINE --runtime 10s

tokio-epoll-uring:

{
  "total": {
    "request_count": 63780,
    "latency_mean": "77ms 993us",
    "latency_percentiles": {
      "p95": "120ms 703us",
      "p99": "143ms 743us",
      "p99.9": "171ms 775us",
      "p99.99": "195ms 583us"
    }
  }
}

Does ca 85-90k IOPS to the NVMe.

std-fs

{
  "total": {
    "request_count": 49303,
    "latency_mean": "100ms 669us",
    "latency_percentiles": {
      "p95": "214ms 399us",
      "p99": "268ms 799us",
      "p99.9": "335ms 359us",
      "p99.99": "399ms 615us"
    }
  }
}

Does ca 70k IOPS to the NVMe.

with higher memroy pre

...

This commit is contained in:

Christian Schwarz

2024-01-29 10:50:52 +00:00

parent 8d6ce71b29

commit 62a3d87098

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

Diff Content Not Available