Skip to content

Performance Benchmarks

Using Intel HiBench on a 4C8G single-node environment (), we compare the business performance overhead before and after loading telemetry components. Data scale: HiBench small profile (WordCount 320MB, SQL 100K rows, KMeans 3M samples). Each workload runs a baseline first (without telemetry), then a with-telemetry run, verifying that metrics reach MySQL.

Spark 3.2.0 + Hadoop 3.2.0

Omnipackage loaded via spark.plugins, spark.telemetry.otel.export.interval.ms=5000, all metric categories enabled.

Workload Baseline Telemetry Overhead Metrics Arrived
micro/wordcount 14.9s 18.5s +24.7% YES
micro/sort 14.2s 12.6s -11.8% YES
micro/terasort 20.1s 19.2s -4.4% YES
micro/repartition 15.9s 14.2s -10.8% YES
sql/aggregation 22.8s 20.2s -11.5% YES
sql/join 24.3s 25.2s +3.7% YES
sql/scan 22.1s 23.5s +6.3% YES
ml/kmeans 29.4s 29.4s -0.1% YES
ml/lr 72.5s 75.7s +4.4% YES
websearch/pagerank 17.4s 15.1s -13.3% YES

Conclusion: All 10 workloads passed, all metrics verified to have arrived in MySQL. Average overhead approximately -1.3% (within measurement noise).

MR Agent + Hadoop 3.2.0

MR Agent injected via -javaagent into mapreduce.map/reduce.java.opts.

Workload Baseline Telemetry Overhead Metrics Arrived
micro/wordcount 50.6s 27.5s -45.7% YES
micro/sort 41.3s 25.7s -37.8% YES
micro/terasort 45.6s 28.2s -38.1% YES

Conclusion: All 3 workloads passed. Telemetry runs are faster due to differences in MR task count and JVM warmup, not agent effects. All mr.task.* metrics verified to have arrived in MySQL.

Hive Hook + Hadoop 3.2.0

Hive Hook injected via hive.exec.post.hooks. Tested with Hive 3.1.3 and 2.3.9, both using the MR engine.

Hive 3.1.3

Workload Baseline Telemetry Overhead Metrics Arrived
sql/aggregation 55.8s 56.8s +1.7% YES
sql/join 98.9s 99.3s +0.4% YES
sql/scan 62.4s 62.9s +0.8% YES

Hive 2.3.9

Workload Baseline Telemetry Overhead Metrics Arrived
sql/aggregation 55.7s 52.5s -5.7% YES
sql/join 97.2s 97.9s +0.7% YES
sql/scan 61.6s 61.9s +0.4% YES

Conclusion: All 12 Hive runs succeeded, all hive.query.* metrics verified to have arrived in MySQL. Hook overhead <2%.

Compatibility Matrix

Component Hadoop 2.7.0 Hadoop 3.2.0 Spark 2.4.4 Spark 3.2.0 Hive 2.3.9 Hive 3.1.3
Spark Plugin (Omnipackage) - PASS - PASS - -
MR Agent - PASS - - - -
Hive Hook - PASS - - PASS PASS

Test Environment

  • Hardware: 4C8G single node (), Java 8 (/opt/jdk8u482-b08)
  • Data Flow: Plugin/Agent/Hook -> OTLP gRPC -> OTel Collector -> Kafka -> Flink Consumer -> MySQL
  • HiBench Version: 8.0-SNAPSHOT, small profile
  • Benchmark Script: benchmark/auto_bench.sh