How fast is it, really? Let’s find out and keep track.

We benchmark LavinMQ to see how it performs under pressure. This page gives you the latest numbers, previous runs, our full test setup, and some honest thoughts on the limitations. The goal is to be transparent. So you can form your own opinion.

v2.6.0

Latest benchmark

This is the latest snapshot of LavinMQ’s performance.

Burstable general-purpose

t4g: Uses ARM based AWS Graviton2 processors and are a cost-effective, burstable general-purpose instance type.

Message size in bytes: 16

Instance type

t4g.micro

Avg. Publish rate

578,788 msgs/s

Avg. Consume rate

578,598 msgs/s

Comment

2 vCPU, 1 GiB RAM

Memory optimized

r7g: Uses ARM based AWS Graviton3 process are optimized for memory intensive workloads.

Message size in bytes: 16

Instance type

r7g.medium

Avg. Publish rate

860,652 msgs/s

Avg. Consume rate

860,411 msgs/s

Comment

1 vCPU, 8 GiB RAM

CPU optimized

c8g: Uses ARM based AWS Graviton4 and are ideal for compute-intensive workloads, such as high performance computing.

Message size in bytes: 16

Instance type

c8g.large

Avg. Publish rate

1,007,903 msgs/s

Avg. Consume rate

1,007,687 msgs/s

Comment

2 vCPU, 4 GiB RAM

Storage optimized

i7i, i7ie: Uses AMD based 5th generation Intel Xeon Scalable and are optimized for storage-intensive workloads that require high-speed access to data residing on local NVMe storage.

Message size in bytes: 16

Instance type

i7i.large

Avg. Publish rate

760,731 msgs/s

Avg. Consume rate

760,645 msgs/s

Comment

2 vCPU, 16 GiB RAM

Transparent setup. Reproducible code.

How we run the tests

We use open-source tools and custom scripts to run our benchmarks. Nothing hidden or proprietary. The full setup is on GitHub, including:

  • Test scenarios
  • Infrastructure configuration
  • Data collection and visualization

Some honest criticism of our approach

What These Numbers Don’t Show

No benchmark is perfect. Here are some caveats to keep in mind:

  • Synthetic tests don’t reflect every real-world workload
  • Network and disk performance can skew results
  • Benchmarks are snapshots, not full stories

Got a benchmark of your own?

We’d love to see it.

If you’ve done your own testing of LavinMQ (or compared it to other brokers), let us know. We may link to third-party benchmarks here in the future.

Contact us