1/4
Your IOPS numbers don’t match the spec sheet? Here’s the performance-testing checklist I actually follow
  

George Fady Lv2  Posted 2026-Oct-27 20:16

I am George Fady, I am a technical support engineer in Sangfor.

“We’re not getting the IOPS we were promised” is one of the most common escalations I get, and most of the time the issue isn’t the storage - it’s how the test was run. Before I accept a performance number as valid, I check:

  • What’s actually being measured - IOPS evaluates small-block IO performance and latency/concurrency, while throughput (MB/s) evaluates large-block IO and bandwidth. A test that reports a “low” number might just be measuring the wrong thing for the workload.

  • Block size - IOPS figures are normally quoted at 4K blocks. Comparing a 4K IOPS number against a test run at a different block size isn’t a fair comparison, and it’s the #1 reason numbers “don’t match the spec sheet.”

  • Tooling - I ask whether the test used IOMeter or FIO, and whether it was configured the standardized way or just run with default settings.

  • Test conditions - was this run against a loaded production cluster, or an idle one? Concurrent production IO will change the result every time.

  • Precautions - several performance-testing precautions exist specifically because a badly configured test can itself degrade production performance while it runs, so I confirm the test was scoped safely before trusting its output.


Nine times out of ten, walking through this list explains the “gap” before we ever touch the storage layer. Question for the community: when you get a “performance doesn’t match spec” ticket, what’s the first thing you ask the customer to confirm?

Like this topic? Like it or reward the author.

Creating a topic earns you 5 coins. A featured or excellent topic earns you more coins. What is Coin?

Enter your mobile phone number and company name for better service. Go