Talk about your next project
ScyllaDB content marketing

The lesson

Publish a competitive benchmark with the headline findings first and the full test setup, tuning and raw results after, so a skeptical engineer can audit the claim instead of taking it on trust.

The move

The hero states the matchup as the title, '4 Nodes of ScyllaDB on AWS i3.metal vs 40 nodes of Cassandra on i3.4xlarge', with the subtitle 'Performance Report: Getting the Most out of AWS i3 Bare Metal Instances'. A short intro leads to a 'Summary of Findings' list such as '2.5X reduction in AWS EC2 Bare Metal costs' and 'Up to 11X improvement in 99th percentile latency'. 'Test Methodology and Configuration' follows, with sections like 'Why Bare Metal?' and 'Why a 40-node Cassandra Cluster?', a side-by-side setup table, and the actual cassandra.yaml, jvm.options and IO tuning settings. 'Benchmark Test Results for ScyllaDB Open Source 2.2 vs Apache Cassandra 3.11 on AWS i3' then opens with a raw data table for each workload.

Why it works

  • Our readThe findings list gives a buyer the outcome in a few lines before any methodology, so the claim lands even if they stop reading early.
  • Our readSections like 'Why Bare Metal?' and 'Why a 40-node Cassandra Cluster?' answer the fairness objection a reader would raise about comparing different instance types and cluster sizes.
  • Our readPrinting the competitor's tuning settings and inviting readers to suggest other settings shows the test can be checked, which matters to an engineering audience wary of vendor benchmarks.

What not to copy

Don't copy ScyllaDB's numbers, hardware choices or its architecture argument against the JVM. Keep the order of matchup headline, findings summary, explained methodology, exposed configs and raw results tables.

More from ScyllaDB

More content marketing

Steal it for your brand.

Tell us what you want to make. We’ll talk through how to put these ideas to work for your brand.