Enterprise Multi‑Data Center Evolution: From Two‑Region Three‑Center to Distributed Active/Active Architecture

After the massive centralization of data, enterprises' business activities increasingly rely on data centers and networks as IT infrastructure, and 7×24 continuous operation becomes the goal of large‑enterprise IT construction, operation, and business. Reducing or eliminating the impact of normal and abnormal downtime on business availability is a key objective for IT teams and a concern for enterprise decision‑makers.

For disaster‑recovery purposes, enterprises typically build two or more data centers (see Figure 1). The primary data center handles core user services, while other centers handle non‑critical services and back up the primary’s data, configuration, and applications. Under normal conditions each center performs its role; when a disaster strikes, the backup can quickly restore data and applications, reducing user loss. Because disasters are low‑probability events, the primary‑backup model only activates the backup during disasters, and as disaster‑recovery standards (GB/T 20988‑2007) rise, more backup IT resources and funds are invested without direct reuse, causing waste. Moreover, the primary‑backup model requires long switchover times and complex relationships, often severely affecting user transactions.

Typical high‑end users such as banks adopt a “two‑region three‑center” solution (production data center, same‑city disaster‑recovery center, remote disaster‑recovery center). In this model multiple data centers have primary‑secondary relationships with differing deployment priorities; disaster response and switchover periods are long, making RTO and RPO goals unattainable, resource utilization low, and ROI unsatisfactory. The approach essentially stacks resources to improve availability, yielding only quantitative changes without a substantive leap in business continuity.

Currently, many industry users—including banks, governments, public transport, and energy—are shifting focus to Distributed Active/Active Data Centers (see Figure 2). Distributed active/active spreads workloads across multiple data centers that cooperate to serve customers. It features two key aspects—distribution and active‑active—reflecting new thinking on resource scheduling, utilization, and flexible business deployment for enterprise‑level users.

Distribution means that data centers are physically and logically spread across facilities, geography, and compute/storage/network resources, satisfying disaster‑recovery and business‑linkage requirements. Multiple DCs can be built incrementally, maintain independence, and be upgraded compatibly; resource scheduling can span multiple DCs, enabling global operation and resource sharing, effectively forming a single large data center.

Active‑active means all centers have equal status, collaboratively providing services under normal operation, fully utilizing resources and avoiding idle backup sites, thus delivering double or multiple times the service capacity of primary‑backup models. When one data center fails, others continue operating and can take over critical or all workloads, achieving fault‑transparent service.

In practice, users often focus on dual‑active data centers; the term “dual‑active” appears more frequently than “active‑active” in technical documents. Technically, dual‑active is a simplified subset of active‑active and the most common model; many dual‑active solutions are derived from active‑active designs, though some may not scale to full active‑active scenarios. Dual‑active is therefore a necessary stage toward active‑active.

While the benefits of active‑active data centers are evident, their construction is a complex system‑engineering effort (see Figure 3). Technically it involves server/VM clustering, data replication and synchronization, and especially cross‑data‑center network interconnection and user access, making network performance and traffic management critical. Network access‑control migration, gateway and data‑sync bandwidth and QoS requirements, IP addressing, routing, gateway design, firewall session handling, traffic‑path planning, and detour control are all design challenges that must be addressed.

Beyond technology, active‑active data‑center construction also encompasses disaster‑recovery, distributed application deployment, process re‑engineering (e.g., DNS, GSLB L4‑7), and cross‑center operational collaboration, imposing higher demands on personnel organization and process design.

Conclusion: Distributed active/active data centers share similarities with cloud‑computing architectures but also differ. Cloud formation can be based on distributed data‑center technology, and the distributed active/active model has a lower implementation threshold. Enterprises focusing on business connectivity, rapid response, and continuous IT optimization can achieve better support for complex applications without blindly chasing advanced solutions, thereby ensuring steady progress of large‑scale data‑center operations.