Dynamic evolution of distributed component-based systems (DCS) is an important task in software engineering. Several challenges are posed in this process. For example, how to preserve consistency during evolution and how to reflect the abstract evolution specification in the concrete reconfiguration implementation. Having observed the generality of software architecture, researchers have proposed various architectural description languages (ADLs), enabling evolution techniques, etc. to investigate the problem. These approaches typically employ the formal semantics of dynamic ADLs at the incremental levels of refinement in the design phase or the explicit maintenance of software architecture at runtime. However, different ADLs usually address different concerns and the lack of runtime support for the causal relation between ADLs and the running system easily leads to the mismatch between them, thus inevitably sacrifices their usability. We propose an approach based on a runtime architecture which is visually generated from an attributed type graph meta-model, exists through the lifecycle of DCS, establishes the causal relation between architectural topology and system configuration, and directs the dynamic evolution.