Overview

This section provides comprehensive performance information for Rumi, including benchmarking results, test methodologies, and performance analysis.

Performance Documentation

Rumi performance documentation is organized into two main sections:

1. Canonical Benchmark

The Canonical Benchmark section presents results from the official end-to-end performance benchmark used to measure Rumi's core capabilities.

What it measures: Complete Receive-Process-Send flow of a clustered microservice, including messaging, state management, persistence, cluster replication, and consensus protocol.

Key metrics:

• Wire-to-Wire Latency: Time from inbound message arrival to outbound message transmission (50th, 99th, 99.9th percentiles)

• Maximum Throughput: Messages processed per second under saturated load

Test program: ESProcessor from the Rumi Performance Benchmark Suite

Results available: Performance metrics for Rumi 4.0 releases

➡️ View Canonical Benchmark Results

2. Performance Benchmark Suite

The Performance Benchmark Suite section documents the collection of benchmarking tools that measure individual Rumi runtime components.

What it includes: 7 modules, each focusing on a specific component:

• Time Module: Time API overhead

• Serialization Module: Message encoding/decoding performance

• Link Module: Cluster replication transport

• Messaging Module: Pub/sub messaging layer

• Persistence Module: Message and data persistence

• Storage Module: Object store operations

• AEP Module: End-to-end canonical benchmark (described above)

Purpose: Isolate and measure individual component performance, understand performance characteristics, validate configurations

Source code: github.com/neeveresearch/nvx-rumi-perf

➡️ Explore Benchmark Suite

Quick Start

View Latest Results

See the latest canonical benchmark results:

• Rumi 4.0.4 Performance - Latest tested release

Run Your Own Tests

Download and run the benchmark suite:

1. Download distribution from the Neeve artifact repository:

   wget https://nexus.n5corp.com/repository/maven-public/com/neeve/nvx-rumi-perf-dist/4.0.4/nvx-rumi-perf-dist-4.0.4-linux-x86-64.tar.gz

2. Extract and run:

   tar xvf nvx-rumi-perf-dist-4.0.4-linux-x86-64.tar.gz
   cd nvx-rumi-perf-4.0.4
   $JAVA_HOME/bin/java -cp "libs/*" com.neeve.perf.aep.engine.ESProcessor --count 10000 --rate 5000

3. See documentation for detailed configuration and parameters:

   • Test Description - Complete methodology

   • Benchmark Suite Documentation - All modules and tools

Understanding Performance

Latency Characteristics

Rumi is optimized for ultra-low latency:

• Typical latency: 27-30µs median (wire-to-wire, including network)

• Tail latency: 99.9th percentile within 1.5x of median

• Configuration: Performance varies by CPU configuration and optimization mode

Throughput Characteristics

Rumi supports high-volume scenarios:

• Typical throughput: 280K+ messages/second per microservice instance

• Scaling: Linear scaling with message complexity and CPU resources

• Configuration: Best throughput with minimal CPU configuration for lightweight handlers

Performance Factors

Key factors affecting Rumi performance:

1. Message Access Method: Direct access (serializer/deserializer) vs Indirect (POJO) - ~10% latency difference, 2.4x throughput difference

2. CPU Configuration: MinCPU, Default, or MaxCPU - affects parallelization vs coordination overhead

3. Optimization Mode: Latency or Throughput - different JVM and runtime tuning

4. Hardware: CPU, memory, storage (NVME vs SSD), network (InfiniBand vs Ethernet)

5. Network Tuning: VMA, RDMA enablement (not enabled in baseline tests)

Next Steps

• Canonical Benchmark - View official end-to-end performance results

• Benchmark Suite - Explore component-level benchmarks

• Test Description - Understand test methodology in detail

• Test Results - Browse results by Rumi release