How to simulate an architecture before you build it
Describe a workload, run it against the modelled architecture, and read what the analysis says about where the design strains.
When this is useful
- Two designs are on the table and the argument for each is currently an opinion.
- A launch volume is known and nobody has checked whether the design carries it.
- You want to find the obvious weaknesses before the load test, not during it.
Before you start
- An architecture modelled with its components and connections. A design starting from one of the simulation architecture templates is a fast way to get to something runnable.
Step 1: Set up the simulation
The Guided Setup Wizard walks the configuration; the Simulation Panel is where the run itself lives.
Simulate ▸ Guided Setup WizardSimulate ▸ Simulation Panel
Step 2: Describe the workload
Workload Profile is where the demand is stated. This is the input that determines everything downstream, so it is worth arguing about the numbers before running anything.
Simulate ▸ Workload ProfileSimulate ▸ Industry Pack
Step 3: Run the analysis
Sensitivity Analysis shows which assumptions the outcome actually depends on. Chaos Engineering explores what failure does to it. Resilience Score summarises, and Recommendations proposes changes.
Metrics & Playback replays the run, which is the form worth showing other people.
Simulate ▸ Sensitivity AnalysisSimulate ▸ Chaos EngineeringSimulate ▸ Resilience ScoreSimulate ▸ RecommendationsSimulate ▸ Metrics & Playback
Step 4: Read the result for what it is
These are model-based estimates derived from the architecture you described and the workload you supplied. They are good for comparing designs and finding obvious weaknesses. They are not measurements of a running system and do not replace load testing.
What happens next
Where the analysis identifies a weakness, record it as a risk or an NFR against the component so the finding outlives the simulation run.
Example
Two queue topologies compared under the same workload profile: sensitivity analysis showed the outcome turned almost entirely on consumer scaling rather than on the topology, which ended the debate the team had been having for a fortnight.
Tips
- Spend the time on the workload profile. Everything downstream inherits its assumptions, and a precise result from a guessed input is still a guess.
- Use sensitivity analysis early. Knowing which assumption dominates tells you which number to go and verify.
Limitations
- Simulation reasons about the model, not the system. It cannot know about behaviour you did not describe.
- Simulation capabilities are separately enabled.