Advantages, Disadvantages, and Principles of Layered Architecture

The article begins by noting the common belief in a "silver bullet" for internet technology and observes that layering appears in many design contexts such as micro‑service architectures, data‑warehouse schemas, and protocol specifications, prompting a discussion of its universal advantages, disadvantages, and guiding principles.

Advantages of layering are summarized as five points: abstract stability, functional reuse, cohesion, hiding complexity and change, and scalability. An early example from a QQ Space WNS framework shows how an acc layer, a dispatch layer, and a webapp layer work together, illustrating how the dispatch layer hides routing complexity and enables fault‑tolerant traffic steering.

The article also lists typical scenarios where layering provides value, such as protocol abstraction (ARP/DNS), routing proxies, storage access, and shielding IP/MAC address changes. It further discusses how multi‑level paging and read/write diffusion in large‑scale systems benefit from layered designs.

Disadvantages are grouped into three categories: increased system complexity, performance/storage overhead, and added dependencies. Over‑layering can make a system harder to understand, introduce latency in high‑throughput paths, and reduce overall availability because each added layer compounds failure probability.

Design Principles are then presented: retain layers that are essential for global consistency, allow cross‑layer calls but forbid reverse calls, prefer dependencies on lower layers, keep lower layers stable, permit multiple sub‑layers within a layer, and avoid creating layers solely for future, uncertain requirements.

An example from data‑warehouse design illustrates the typical layers dwd (cleaned raw data), dwm (light aggregation), dm (subject‑oriented tables), and app (high‑level reports). The discussion highlights when a dim layer is necessary and when it can be omitted.

The article concludes with a concise summary: layering brings abstract stability, reuse, cohesion, complexity shielding, and scalability, but also adds complexity, performance cost, and dependency risk; therefore, layers should be kept only when their benefits outweigh the drawbacks, following the stated principles.