results: req_lru_size=2 gives tokio-epoll-uring 16k GetPage/s@110kIOPs std-fs: 9.5 GetPage/s @ 65k IOPS

Browse Source

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   --limit-to-first-n-targets 1000 --set-io-engine tokio-epoll-uring --set-req-lru-size 2 --runtime 2m

Biggest gain with lru_size from 0 to 1, yay.
Adding one more gives another 1-2k

cgroup mem.high unlimited
made sure global page cache is large enough to not have any misses

MAKE SURE TO WARM UP, IT TAKES A WHILE, STILL DON'T KNOW WHY WARMUP IS
THAT BADLY NEEDED

std-fs: 50% cpu, lot of iowait
2024-01-29T18:25:52.923572Z  INFO all clients stopped
{
  "total": {
    "request_count": 1194213,
    "latency_mean": "68ms 343us",
    "latency_percentiles": {
      "p95": "152ms 63us",
      "p99": "201ms 215us",
      "p99.9": "260ms 991us",
      "p99.99": "314ms 623us"
    }
  }
}

tokio-epoll-uring: 100%cpu utilization
Disk isn't saturated.
We're CPU bound here.

{
  "total": {
    "request_count": 1927700,
    "latency_mean": "43ms 11us",
    "latency_percentiles": {
      "p95": "83ms 263us",
      "p99": "101ms 887us",
      "p99.9": "124ms 991us",
      "p99.99": "147ms 583us"
    }
  }
}

...

This commit is contained in:

Christian Schwarz

2024-01-29 18:22:18 +00:00

parent 49a5e411d6

commit 6753ff089c

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