Log-Parser LogHub-2k Benchmark Report

Generated: 2026-07-31T01:26:59.678Z · ← back to benchmarks · LogHub Full →

16
Datasets
25
LLM Calls
88K
Tokens
$0.123
Est. Cost

Dataset Comparison — drain-ts vs log-parser

Datasetdrain-ts GAdrain-ts PTA LP Drain GALP Drain PTA LP Refined GALP Refined PTA GA ΔPTA Δ
HDFS 99.9%76.2% 99.9%76.2% 99.9%76.2% ~ ~
Hadoop 99.9%79.7% 99.9%79.8% 99.9%79.8% ~ ~
Spark 100.0%89.8% 100.0%89.8% 100.0%89.8% ~ ~
OpenStack 96.0%73.1% 96.0%73.1% 96.0%75.5% ~ +2.4pp
Zookeeper 99.9%88.8% 99.9%88.8% 99.9%88.8% ~ ~
BGL 100.0%83.1% 100.0%83.1% 100.0%83.8% ~ +0.7pp
HPC 99.8%85.5% 99.8%85.5% 99.8%85.5% ~ ~
Thunderbird 98.0%83.4% 98.0%83.4% 98.0%83.4% ~ ~
Linux 100.0%85.5% 100.0%85.4% 100.0%88.4% ~ +3.0pp
Mac 93.8%79.4% 93.8%79.4% 93.8%79.4% ~ ~
Apache 100.0%92.1% 100.0%92.1% 100.0%92.1% ~ ~
OpenSSH 100.0%81.1% 100.0%81.1% 100.0%83.9% ~ +2.8pp
Windows 99.8%87.8% 99.8%87.8% 99.8%88.6% ~ +0.8pp
Android 99.9%71.7% 99.8%71.7% 99.8%69.1% ~ -2.6pp
HealthApp 100.0%87.9% 100.0%87.9% 100.0%87.9% ~ ~
Proxifier 98.0%77.5% 98.0%75.0% 98.0%91.7% ~ +16.7pp

LLM-enhanced datasets: OpenStack, Thunderbird, Windows, Android, Proxifier · SynLogRefiner applied to all others · skipRefinement: HDFS, Hadoop, Spark, Zookeeper, HPC, Mac, Apache, HealthApp