This is a prerelease version.

View latest

Test Hazelcast performance with Simulator

5.7-SNAPSHOT 5.5 5.4 5.3 5.2 5.1 5.0

Hazelcast Simulator is a high-performance, production-grade testing framework for running performance, stress, and latency tests on Hazelcast clusters. It allows you to simulate complex workloads and evaluate distributed systems in a realistic and reproducible manner.

Hazelcast Simulator is designed for:

  • Validating new features.

  • Detecting regressions.

  • Measuring throughput and latency under varying loads.

  • Simulating real-world failures (such as network latency and node crashes).

  • Running tests against cloud-deployed Hazelcast clusters.

Key capabilities

  • Supports both throughput and latency-oriented testing modes.

  • Orchestrates tests across multiple clients and members, with configurable topologies.

  • Provides out-of-the-box support for static infrastructure.

  • Support for automatic provisioning of cloud infrastructure (currently for AWS).

  • Integrates with monitoring tools and profilers for system-level analysis.

  • Enables custom test logic with a flexible Java-based test API.

Supported test types

Simulator enables the execution of a variety of performance test types:

  • Load Tests: Combined throughput and latency under realistic load.

  • Max Throughput Tests: Identify system saturation points under increasing concurrency.

  • Latency Tests: Measure operation response time at fixed request rates.

  • Spike and Soak Tests: Evaluate short bursts and long-term stability.

  • Stress Tests: Push the system until failure to observe limits.

Performance testing strategy

Before executing tests:

  • Use a test plan to define test scope, goals, and configurations.

  • Choose appropriate machine types, network configuration, and topology.

  • Consider persistence, CPU, memory, and network bandwidth requirements.

Executing tests:

  • Start with a small cluster and progressively scale load by adjusting parameters such as threadCount and ratePerSecond.

  • Continuously monitor throughput (TPS), latency percentiles, CPU utilization, memory consumption, and other relevant system metrics.

  • Use Hazelcast Simulator’s latency measurement and ramp-up utilities to apply load in a controlled and reproducible manner.

After executing tests:

  • Analyze the collected metrics and logs to detect bottlenecks, stability issues, and scaling thresholds.

  • Compare observed results against the goals defined in the test plan and identify bottlenecks or deviations from the planned outcomes.

  • Conclude by shutting down or cleaning up all cluster instances and related resources to avoid interference with subsequent runs.

Advanced features

  • Network Latency Simulation: Inject delays between groups of machines.

  • CP Subsystem Testing: Configure CP member priorities using cp_priorities.

  • Flight Recorder Integration: Enable JFR for profiling with member_args.

  • Warmup/Cooldown: Configure warmup and cooldown periods to improve report accuracy.

lifebuoy

Help and support