For the purposes of this test we used a separate environment that mirrors our production environment. We took this environment through 3 steps of scale: from 1 node for each relevant service to 3 and then 5 nodes.
For each scale step we load tested the Awell Orchestration API up to the point the API response time surpassed 500ms - as our target is to always offer under 500ms response time.
This is a the base of the test where we deployed only one instance of each of our services.
As expected the system reached it limit relatively fast, at 150 requests per second. From this point on the system response times spike to an average of 800ms with requests taking up to 2000ms to get a response.
By default, our services are running on 3 nodes.
By increasing the deployment to 3 nodes of each service we already see an improvement of the performance.
At 3 nodes the system can already process more than 200 requests per second with response times well under 100ms.
As expected, we see an increase of the response times with 200-250 requests per second, averaging at 600ms with spikes up to 2500ms.
Finally, to prove the horizontal scalability of the Awell system we tested against at 5 node deployment.
As observable at 400 requests per second, the system is still responding in average well under 200ms and even the spikes are averaged around 500ms.
This concludes that Awell Orchestration API is scaling horizontally and can guarantee under 500ms response time in average.