Use cases

Idea is to run a continuous or periodic optimization cycle. Ideally, there should be a breeding sub landscape, where the GEAs can do their optimiziation part in a safe, non-invasive or even disruptive way. Replicating load or traffic patterns in such a replicate might appear challenging, but may not be necessary. A proper enough approximation might suffice for the larger fraction of practical use cases.

The optimizer core is working on DEAP workers parallelized with help of a shim layer over commonplace distributed system orchestrator solutions. Further, the feed-back control loop is driven in jobs, perceiving aggregated telemetry and data of similar character gathered from the targeted objects. Also direct inspection of the breeding status has to be foreseen. After running the optimization via parallelized GEAs, the outcome will be applied on the breeder. This cycle will run until a customer defined acceptance or approximation epsilon has been reached or an another, maybe time-bound termination kicks in.

Eventual outcome shall be a configuration perceived sufficiently optimal. It must be handed in structured way to the consumer, allowing him to apply the breeding result at higher criticality or the actual optimization target at his or her arbitration in the way most desired or applicable.

Alternatively, customers will not be hampered from doing all the previously outlined directly on productive stacks. If the targeted sub-components are not of utmost criticallity, that could be a frequent scenario to expect also.

It's a special case of the previously described anyways.