A modern large scale service architecture depends on microservices which connect through an enterprise service bus. Each call depends on redundant items of a resilient infrastructure, such as transient memory and network cables. Despite high availability, these suffer from degradation below noise thresholds but may still impact service quality. The choice of load balancing over the infrastructure items affects the number of items that need to work for service delivery. Operational examples do illustrate.

Operational experience triggered theoretical analysis that revealed an exponential of exponential decay of service resiliency with the number of called microservices for random load balancing over infrastructure. Sticky load balancing keeps the impact linear, because successful traversal in the previous call confirms its availability for the next.

The proposed affinity increases system load imbalance as side effect, so that mixed approaches to control a compromise are introduced.

Advanced statistical math