DeFog is an innovative microservice placement and load balancing approach for distributed multi-cluster cloud-fog-edge architectures to minimise application response times. The architecture is modelled as a three-layered hierarchy. Each layer consists of one or more clusters of machines, with resource constraints increasing towards lower layers. Applications are modelled as service oriented architectures (SOA) comprising multiple interconnected microservices. As many applications can be run simultaneously, and as the resources of the edge and the fog are limited, choosing among services to run on the edge or the fog is the problem this work is dealing with. DeFog focuses on dynamic (i.e., adaptive) decentralised service placement within each cluster with zero downtime, eliminating the need for coordination between clusters. To assess the effectiveness of DeFog, two realistic applications based on microservices are deployed, and several placement policies are tested to select the one that reduces application latency. Least frequently used (LFU) is the reference service placement strategy. The experimental results reveal that a replacement policy that uses individual microservice latency as the crucial factor affecting service placement outperformed LFU by at least 10% in application response time.