Shenzhen Metro Rail Transit ACC Phase II Project

Construction of ACC, CLC Phase II, Internet Ticketing Management System, and Simulation Test Cloud Platform System for Shenzhen Metro. Used for: Line 12, 33 stations, maximum daily ridership of 389,900; Line 14, 18 stations, maximum daily ridership of 354,200; Line 16, 24 stations, maximum daily ridership of 134,100; Branch Line 6, 4 stations.

As planned, in 2023 it will be connected to the entire Shenzhen Metro network, comprising 16 lines, 312 stations, covering 547.12 km, with a maximum daily ridership of 8,487,300.

Currently, the "Industrial Control Cloud" (Huawei Cloud) in the Shenyun Center comprises bare-metal servers (mainly for database servers) and cloud servers (mainly for business platforms). Servers at each station along the lines (combined database and business) interact with the Shenyun Center through business operations. The project is built on GBase 8s database from General Data Technology (GBase).

• Database Architecture

CLC Phase II and the Internet Ticketing Management System deployed SSC and HAC architectures at the primary production cloud center and each site to enable high-availability clustering, and underwent comprehensive application adaptation and verification tests covering connectivity, destructive testing, stability, and performance. After successfully passing the tests, the system was officially launched and went live.

• Advantages of GBase 8s High-Availability Architecture

SSC Architecture: Multiple nodes can access shared storage simultaneously; memory synchronization between nodes is achieved through LSN transmission; secondary nodes support read and write operations (with the final write step routed to the primary node); cluster secondary nodes can be dynamically added or removed, scaling up to 128 nodes; load balancing across nodes and read/write separation are supported.

HAC Architecture: Data replication at the database instance level; standby nodes are readable and writable, enabling read/write separation; used for high availability within the same data center or same-city disaster recovery; flexible modes based on RPO: fully synchronous mode, asynchronous mode, and semi-synchronous mode.

Since go-live, the GBase 8s database has delivered exceptional performance that fully meets the business demands.

• High Concurrency: Supports 30,000 concurrent user connections; sustains 4,000 concurrent users stably for over one hour.

• Efficient Backup & Recovery: Full backup speed reaches 170 MB/s for 1 GB of data, and complete recovery reaches 22 MB/s.

• Data loading speed of 1 GB text data reaches 200 MB/s, and data export speed reaches 390 MB/s.

• Efficient Query Performance: For tens of millions of records, a single-table range query on an indexed field with one condition delivers an average response time in milliseconds.

• For hundreds of millions of records, the same type of query delivers an average response time in seconds.

• For tens of millions of records, 50 concurrent combined queries across multiple indexed fields achieve an average response time in seconds.

• On hundreds of millions of records, full-pattern fuzzy search retrieval time is under 0.7 ms, leading-pattern fuzzy search under 0.6 ms, and trailing-pattern fuzzy search under 0.4 ms.

• Efficient Data Processing: A single table can store up to one billion records.

• Sustained insertion rate exceeds 5,000 records per second per table, with batch import speed reaching 120,000 records per second.

• On a dataset of tens of millions of records, single-row insertion takes 3 ms, updating a row takes 2 ms, and a 10,000-row batch update completes in 90 ms.

• Efficient Transaction Processing: In a TPC-C benchmark with 500 warehouses, TPMC score reached 1.6 million.