A team picks a workflow tool. It runs well in testing, so they ship it.
A year later, traffic is up. The bill is up even more.
Nobody planned for that, because nobody asked the right question at the start. Not "how fast is it?" but "what will it cost when we're ten times bigger?"
Turn speed into daily volume
Throughput numbers are usually quoted per second, which makes them hard to feel. A quick conversion helps: multiply transactions per second by 86,400, the number of seconds in a day.
- 1,000 per second is about 86 million a day
- 10,000 per second is about 864 million a day
Now compare that with what you really expect. A lot of teams find they need far less than they feared. Others find they need a lot more than they planned for. Either way, you can't judge a tool until you know your own number.
Watch how the cost grows
A tool can be fine at small volume and painful at large volume. Here's an illustration. Spending $20,000 a year for a few thousand transactions per second might be reasonable. If handling a bigger load pushed that to $200,000, the math changes. Those are example figures, not anyone's price list, but the shape is real: what matters is whether cost rises in step with traffic or faster.
Simple workflows that run at low volume can live happily with a modest setup. Complex, high-volume ones can't. And a setup that's fast but very expensive tends to be dropped once the bill gets attention.
When latency is the whole point
Some use cases don't give you time to wait. Payment processing, IoT data pipelines and AI-driven workflows are all situations where slow responses or low throughput cause real problems for users. If yours looks like that, test under load before you commit, not after.
What to ask before you choose
- How many transactions per second can one instance sustain, and what did that test look like?
- What happens when you need more? Do you add instances, and how does the price change?
- Does the number come from a workflow like yours, or a simple one-step example?
- What does it cost at your expected volume, and at ten times that?
Those four questions tell you more than any single benchmark.
For reference, Unmeshed reports up to 10,000 transactions per second on a single instance, which works out to nearly a billion a day, and says larger loads are handled by adding instances to the cluster. As with any vendor's numbers, test it against your own workflows. You can try Unmeshed and see how it behaves with yours.
Top comments (0)