Analysis of Network Resource Requirements for TiDB Two-Region Three-Center Construction

This topic has been translated from a Chinese forum by GPT and might contain errors.

Original topic: TiDB两地三中心建设的网络资源需求分析

| username: 春风十里

In the TiDB two-city three-center architecture, what kind of dedicated network line is recommended for intra-city and inter-city disaster recovery? Must it be directly connected with optical fiber? If using network cables, what is the maximum acceptable latency? Taking 15 kilometers within the same city and 150 kilometers between cities as examples, please explain the recommended network configuration for the two-city three-center construction.

If you know the approximate network costs, it would be best to recommend them as well, such as the bandwidth of a single line and the approximate cost. Our company is a financial institution, and currently, the data volume of our app is not large, about 1TB. We plan to replace MySQL with TiDB, but the infrastructure requirements still need to be evaluated. Currently, we are using a MySQL replication solution with a 100M dedicated line between the two locations.

| username: dba-kit | Original post link

You can check out this article in the official documentation, which introduces the multi-site solution:

| username: 春风十里 | Original post link

I am aware of this architecture. What I want to know is the estimated network cost for a two-site three-center setup. What are the requirements here?

| username: tidb菜鸟一只 | Original post link

This can be evaluated by TiDB staff, right? The main focus is not on the total data volume, but on the volume of changing data, right?

| username: 春风十里 | Original post link

Single Region Dual AZ Deployment of TiDB | PingCAP Documentation Center

TiDB typically adopts a multi-AZ deployment scheme to ensure high availability and disaster recovery capabilities of the cluster. The multi-AZ deployment scheme includes various deployment modes such as single-region multi-AZ deployment mode and dual-region multi-AZ deployment mode. This article introduces the single-region dual AZ deployment scheme, which deploys two AZs in the same region, offering lower costs while still meeting high availability and disaster recovery requirements. This deployment scheme uses an adaptive synchronization mode, namely Data Replication Auto Synchronous, abbreviated as DR Auto-Sync.

In a single-region dual AZ deployment scheme, the two AZs are usually located in the same city or two adjacent cities (e.g., Beijing and Langfang), within 50 km of each other, with network connection latency between AZs less than 1.5 ms and bandwidth greater than 10 Gbps.

The official requirement of “network connection latency between AZs less than 1.5 ms and bandwidth greater than 10 Gbps” is actually quite high. I asked my network colleagues, and in Shanghai, this situation would require bare fiber, costing about one million per year. Additionally, the supporting network multi-layer switches are also very expensive. It feels like the real-time strategy for same-city disaster recovery is still quite costly.