ITPUB Exclusive Interview | Pursuing Business Continuity Without Limits: GBase Database High Availability Builds a Data Security Barrier
"Zero data loss, zero business downtime" is the core demand of mission-critical business systems, which places stringent requirements on database high availability. For Chinese databases to carry the weight and deliver for critical business scenarios, they must have robust high availability capabilities and a relatively complete HA solution portfolio to maximize business continuity. So, what are the mainstream database high availability solutions today? How should enterprises choose the right one? And how can business continuity be better ensured?
To address these questions, Cui Zhiwei, General Manager of the GBase 8s Product Division at General Data Technology Co., Ltd., recently gave an interview to the database technology community ITPUB. He stated: The pursuit of business continuity is endless. The current database market in China features a rich array of database HA architectures, and enterprises need to select the appropriate HA solution based on their business scenario requirements.
The Evolution of Database High Availability Architectures
High Availability (HA) refers to a system's ability to remain operational and continue providing services in the face of various failures such as hardware malfunctions, software crashes, and network issues. It aims to minimize system downtime through design, employing methods like redundancy and automatic failover to improve system availability and avoid single points of failure, thus ensuring high stability and continuity.
Early databases lacked dedicated HA architectures, relying mainly on periodic backups and log archiving for availability. Databases easily became single points of failure. At that time, high availability was more about using clustering software at the operating system level to manage hardware and software, including databases. An example is the IBM High Availability Cluster Multi-Processing (HACMP) software, primarily used on the AIX operating system to provide HA and failover capabilities, ensuring the continuous operation of critical business applications.
In the 1990s, with the rise of the internet and increased enterprise data dependency, HA became a key factor in database system design. Databases began introducing redundancy and failover mechanisms to enhance system reliability and availability.
Database HA architectures have continuously evolved alongside storage, network hardware advancements, and changing business needs, progressing from standalone servers to primary-standby clusters, shared-storage clusters, and distributed clusters, offering an increasingly richer set of HA solutions.
Taking the industry leader Oracle as an example, it has successively launched various solutions including Host HA, Active Data Guard (ADG), Real Application Clusters (RAC), and Oracle Golden Gate (OGG). These HA architectures can be used alone or in combination, providing stable and reliable high availability services for the mission-critical core businesses of financial institutions and telecom operators for many years.
For instance, Oracle's primary-standby HA solution, ADG (earlier known as Data Guard), allowed the primary node to handle reads and writes, while the standby node was initially incapable of serving queries. Oracle 11g introduced Active Data Guard (ADG), enabling standby nodes for queries, thus offloading the primary node and further enhancing system availability. Subsequently, ADG offered three data protection modes—Maximum Protection, Maximum Availability, and Maximum Performance—providing users with flexible choices.
The release of Oracle 9i in 2000 marked a significant milestone, introducing Real Application Clusters (RAC) technology. RAC is a classic shared-everything architecture, meaning a shared-storage cluster technology that allows multiple servers to access a single database instance, thereby providing high availability and load balancing. If one node fails, other nodes can continue providing services, ensuring system continuity. As RAC technology matured, an increasing number of core databases in large enterprises adopted it.
In the context of domestic substitution, although the technology stack of Chinese databases has changed, the business requirements for database high availability have not diminished; they are only growing higher. Chinese databases need to reach the same level of HA technology as Oracle, or even achieve breakthroughs, to progressively replace core business systems across various industries.
Currently, most Chinese databases have built their own complete HA architectures, often inspired by Oracle. The GBase 8s database from General Data Technology Co., Ltd., for example, has a complete HA technology stack that mirrors Oracle's to ensure high availability services for business systems.
For example, GBase 8s primary-standby clusters support two cluster modes: HAC and RHAC. HAC is a primary-standby cluster model that can select from fully synchronous, near-synchronous, and asynchronous redo log replication modes based on network transport conditions, mirroring the Maximum Protection, Maximum Availability, and Maximum Performance modes of Oracle ADG. RHAC supports a primary-multiple-standby architecture using an asynchronous transmission mechanism.
The GBase 8s Shared-Storage Cluster (SSC)—comparable to Oracle RAC—achieves node high availability through shared disks. Data is stored in a single copy, supporting multiple writes and multiple reads, effectively utilizing hardware resources and avoiding data redundancy issues. The shared storage supports both disk arrays and distributed storage.
In the field of centralized databases, RAC-like shared-storage clusters present significant technical challenges, often considered a mountaintop. The breakthrough of Chinese databases with RAC-like clusters undoubtedly opens the door to high-end scenarios. However, although many domestic databases feature RAC-like clusters, few support multiple writes and multiple reads; many standby nodes are read-only, resulting in cluster throughput that is less than the processing power of a single machine.
Cui Zhiwei pointed out that all IT technologies must ultimately serve business development. China's database HA architectures have moved from non-existence to existence and are now on the path from existence to excellence. Moving forward, while continuously refining their HA solutions, everyone will strive to reduce primary-standby switchover times, constantly improve business continuity, and increase cluster throughput.
How to Choose the Right High Availability Solution?
Some might argue that if business continuity is so crucial, one should always use the highest-level HA architecture. If budget is no concern, feel free. But in reality, even highly capable financial institutions like banks must weigh their options carefully when building HA solutions—they cannot afford to be capricious because HA solutions are extremely costly. Be it intra-city disaster recovery or remote disaster recovery, data centers, networks, etc., all require significant financial investment.
Mr. Cui explained that different HA solutions have their own characteristics. He recommends that enterprises select HA solutions based on their business scenario needs, combined with their own budget, data center facilities, and network conditions.
For instance, the primary-standby cluster—a common HA solution—is better suited for smaller data volumes and business scenarios where data consistency requirements are not exceptionally stringent.
On one hand, the redundancy in a primary-standby cluster requires storing multiple data copies, incurring additional storage costs. On the other hand, data synchronization and primary-standby switchover after a failure take time. If the data volume reaches tens of TBs, it leads to higher storage costs and a wider switchover time window. Additionally, data synchronization in primary-standby clusters does not easily guarantee strong data consistency between the two nodes; many Chinese databases compensate for this shortcoming using read-write splitting plugins.
Shared-storage clusters effectively solve the storage cost and strong data consistency problems, making them suitable for business scenarios with large data volumes and high requirements for strong data consistency.
Shared-storage clusters achieve high performance and ensure business continuity with minimal hardware and database instances, representing a cost-effective HA solution. However, shared-storage clusters are complex and have high technical barriers, requiring database maintenance personnel and developers with advanced professional skills. Moreover, these clusters have stringent requirements for hardware and database software. To achieve strong data consistency, information synchronization demands very high network bandwidth; for example, Oracle RAC heartbeat networks typically start at 10 Gigabit Ethernet, and Oracle Exadata even uses dedicated 40GB high-speed networks for internal heartbeats.
Cui Zhiwei noted that for general government and enterprise customer scenarios with less data volume, primary-standby cluster HA solutions are mostly chosen. However, for financial sector scenarios demanding strong data consistency, and for large data volume scenarios (data exceeding 10TB, 20TB), shared-storage cluster solutions are preferred.
GBase 8s offers a relatively complete portfolio of HA solutions for enterprises to choose flexibly, such as primary-standby high availability HAC/RHAC, shared-storage cluster high availability SSC, and enterprise real-time data replication ER.
• The GBase 8s HAC cluster is suitable for businesses sensitive to network latency, ideally deployed within the same city or data center. Considering single-server disk capacity limits, HAC is applicable when data volume does not exceed 10TB. For long-distance transmission and remote disaster recovery scenarios, the RHAC cluster can be used. Due to long-distance transmission with lower bandwidth, which increases network latency, RHAC optimizes bandwidth usage through an asynchronous checkpoint mechanism and data compression.
• The GBase 8s Shared-Storage Cluster (SSC), a RAC-like technology, supports shared storage to guarantee strong data consistency. The cluster employs a peer management model for its controllers, supporting up to 16 nodes, all of which are read-write capable. If the primary controller node fails, a secondary node can be promoted to primary controller, ensuring system high availability. For business scenarios with data volumes below 100TB, the shared-storage cluster is the most cost-effective HA solution compared to distributed databases. For data volumes exceeding 100TB, a distributed database might be more suitable than a centralized one.
• The GBase 8s Enterprise Replication (ER) feature (comparable to OGG) provides built-in, table-level real-time data synchronization capabilities. It is mostly applied in data exchange and sharing scenarios, such as real-time data exchange across ministry, provincial, city, and county departments, or data sharing among parallel units like supermarket chains.
Enterprises can deploy primary-standby HA or shared-storage cluster HA independently, or they can combine them to build higher-level HA solutions like a dual-active data center within a city (SSC+HAC) or a geo-redundant "Two Sites, Three Centers" (SSC+HAC+RHAC) architecture.
GBase 8s "Two Sites, Three Centers" Solution
Currently, the GBase 8s HA cluster solution can achieve an RPO of 0 and an RTO of less than 30 seconds. In real business scenarios, primary-standby switchover can be completed in approximately 10-15 seconds.
Leveraging its outstanding capabilities, GBase 8s HA clusters have already served numerous mission-critical core businesses in industries like finance, rail transit, energy, and government, earning the trust of key industry customers.
For example, the State Grid's dispatching cloud platform, with business data exceeding 50TB, adopted the SSC+RHAC HA solution to successfully replace Oracle. Through the construction of a remote disaster recovery solution spanning thousands of kilometers, it achieved second-level data synchronization for a local and remote read-write separated active-active cluster, with the database completely transparent to applications, and the longest continuous operation exceeding 600 days.
For Shenzhen Metro's CLC Phase II and the Internet Ticketing Management System, an SSC+HAC approach was implemented at the production cloud primary center and various stations, achieving HA cluster functionality. It met requirements for 4,000 concurrent users running stably for over an hour, handling tens of millions of data records with millisecond-level query response times.
A city commercial bank in Southwest China adopted the SSC+HAC HA solution for its core banking system. After the database replacement, it has run continuously online for more than 760 days, fully demonstrating the stability of GBase 8s.
Summary: The Endless Pursuit of Business Continuity
Moving forward, GBase 8s will continuously optimize and refine its HA solutions. This includes providing more granular resource control—locking only the tables involved in the affected transactions when the primary node fails, while other tables remain open for business. Additionally, session persistence and transaction persistence capabilities will be introduced, along with supporting data sharding within shared-storage clusters to reduce contention, among other enhancements.
"Our ultimate goal is to guarantee customer business continuity so perfectly that no matter what database failure occurs, the customer perceives nothing," said Cui Zhiwei. General Data Technology Co., Ltd. is committed to continuously building database products that users can trust. More and more real-world user scenarios have hardened GBase 8s to be more stable and reliable. Currently, several hundred sets of its SSC clusters are already in production, and customer recognition gives him great confidence.
In the era of the digital economy, the enterprise pursuit of business continuity is never-ending. However, with limited resources, achieving better outcomes in both high availability and business continuity requires a combined effort from both users and vendors.
"On the user side, leadership and technical teams should define reasonable HA switchover goals, break the rigid link between critical systems and a specific HA capability, and design HA objectives based on the true characteristics of their business systems. On the database vendor side, product HA capabilities should be optimized according to user application scenario requirements, rather than just competing over an unreliable failover time." To conclude with this passage from Mr. Baishan's article, it is hoped that through the joint efforts of users and vendors, system HA capabilities will continuously advance, and business continuity will be better safeguarded.