ceremonyclient/pebble/docs/index.html
Cassandra Heart 2e2a1e4789
v1.2.0 (#31)
2024-01-03 01:31:42 -06:00

167 lines
6.9 KiB
HTML

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no">
<link rel="stylesheet" href="css/app.css" >
<title>Pebble Benchmarks</title>
</head>
<body>
<div class="center rows">
<div class="section">
<div class="columns">
<div class="title">Pebble Benchmarks</div>
<div class="updated">Last updated</div>
</div>
<br/>
<div class="columns">
<div class="overview">
Benchmarks are run nightly using <a class="code"
href="https://github.com/cockroachdb/pebble/blob/master/cmd/pebble/ycsb.go">pebble
bench ycsb</a> on AWS m6id.4xlarge machines equipped with
local SSD storage. The AWS instances show remarkably high
instance to instance performance variability. In order to
smooth out that variability the benchmarks are run multiple
times each (using different instances) and outliers are
excluded.
</div>
<div>
<div class="controls">
<b>Detail:</b>
<a id="readBytes" class="toggle">Bytes Read</a> |
<a id="writeBytes" class="toggle">Bytes Written</a> |
<a id="readAmp" class="toggle">Read Amp</a> |
<a id="writeAmp" class="toggle">Write Amp</a>
</div>
<div class="controls">
<b>Options:</b>
<a id="localMax">Local scale</a>
</div>
</div>
</div>
<hr class="divider"/>
<div class="annotation" data-date="20200614">L0-sublevels and
flush-splits enabled</div>
<div class="annotation" data-date="20200604">Increased
LogWriter free blocks 4-&gt;16</div>
<div class="annotation" data-date="20200706">Began tracking
ycsb/E read-amp</div>
<div class="annotation" data-date="20201022">Level metadata
switched to use a B-Tree</div>
<div class="annotation" data-date="20201222">Enabled
read-triggered compactions</div>
<div class="annotation" data-date="20210113">Readahead
and preallocation bug fixed</div>
<div class="annotation" data-date="20210326">Removed excess
read samples for read-triggered compactions</div>
<div class="annotation" data-date="20210429">Switched to Ubuntu
20.04.2 LTS AMI</div>
<div class="annotation" data-date="20211018">Read compaction fixes</div>
<div class="annotation" data-date="20211109">Bumped benchmark
runtime to 90 minutes</div>
<div class="annotation" data-date="20220330">Data quality issue introduced (YCSB A only)</div>
<div class="annotation" data-date="20220526">Data quality issue fixed (YCSB A only)</div>
<div class="annotation" data-date="20220626">Began zeroing reused
iterator structs (#1822)</div>
<div class="annotation" data-date="20230215">Grandparent boundary
compaction splitting</div>
<div class="annotation" data-date="20230516">Infrastructure
change (#2578)</div>
<div class="annotation" data-date="20230607">ycsb/F sampling bug</div>
<div class="annotation" data-date="20230611">Switched to m6id.4xlarge
(from 5d.4xlarge)</div>
</div>
<div class="section rows">
<div>
<span class="subtitle">YCSB A</span>
<span>(50% reads, 50% updates, zipf key distribution)</span>
</div>
<div class="columns">
<svg class="chart ycsb" data-key="ycsb/A/values=64"></svg>
<svg class="chart ycsb" data-key="ycsb/A/values=1024"></svg>
</div>
</div>
<div class="section rows">
<div>
<span class="subtitle">YCSB B</span>
<span>(95 reads, 5% updates, zipf key distribution)</span>
</div>
<div class="columns">
<svg class="chart ycsb left" data-key="ycsb/B/values=64"></svg>
<svg class="chart ycsb right" data-key="ycsb/B/values=1024"></svg>
</div>
</div>
<div class="section rows">
<div>
<span class="subtitle">YCSB C</span>
<span>(100% reads, zipf key distribution)</span>
</div>
<div class="columns">
<svg class="chart ycsb left" data-key="ycsb/C/values=64"></svg>
<svg class="chart ycsb right" data-key="ycsb/C/values=1024"></svg>
</div>
</div>
<div class="section rows">
<div>
<span class="subtitle">YCSB D</span>
<span>(95% reads, 5% updates, uniform key distribution)</span>
</div>
<div class="columns">
<svg class="chart ycsb left" data-key="ycsb/D/values=64"></svg>
<svg class="chart ycsb right" data-key="ycsb/D/values=1024"></svg>
</div>
</div>
<div class="section rows">
<div>
<span class="subtitle">YCSB E</span>
<span>(95% scans, 5% updates, zipf key distribution)</span>
</div>
<div class="columns">
<svg class="chart ycsb left" data-key="ycsb/E/values=64"></svg>
<svg class="chart ycsb right" data-key="ycsb/E/values=1024"></svg>
</div>
</div>
<div class="section rows">
<div>
<span class="subtitle">Insert-only</span>
<span>(100% inserts, zipf key distribution)</span>
</div>
<div class="columns">
<svg class="chart ycsb left" data-key="ycsb/F/values=64"></svg>
<svg class="chart ycsb right" data-key="ycsb/F/values=1024"></svg>
</div>
</div>
</div>
<div class="center rows">
<div class="section">
<hr class="divider"/>
<div>
<span class="subtitle">Write throughput</span>
<span>(100% inserts, zipf key distribution)</span>
</div>
<br />
<div class="overview">
This benchmark attempts to find the optimal write throughput by
driving more and more load against the DB until a target heuristic
fails (currently a mixture of number of L0 sublevels, L0 files, and
whether the DB has experienced a write stall). These benchmarks are
run nightly using <a class="code"
href="https://github.com/cockroachdb/pebble/blob/master/cmd/pebble/write.go">pebble
bench write</a> on GCP n2-standard-32 machines equipped with 16 local
SSDs in a RAID 0 array.
</div>
</div>
<div class="section rows">
<div class="columns">
<svg class="chart write-throughput" data-key="write/values=1024"></svg>
<svg class="chart write-throughput-detail"/>
</div>
</div>
</div>
<script src="js/d3.v5.min.js"></script>
<script src="https://pebble-benchmarks.s3.amazonaws.com/data.js"></script>
<script src="js/write-throughput.js"></script>
<script src="js/app.js"></script>
</body>
</html>