Bank of Nanjing
Company profile
Bank of Nanjing was founded on February 8, 1996. It is a joint-stock commercial bank with independent legal entity status. Its shares are held by the state, Chinese corporations, foreign investors, and private individuals. In 2001 and 2005, the International Finance Corporation (IFC) and BNP Paribas became shareholders. It was the first city commercial bank in China to begin the pre-listing advisory process and was successfully listed in 2007. In 2017, the bank was ranked 146th in The Banker's Top 1000 World Banks and 131st in the Top 500 Banking Brands. To keep pace with the rapid growth of Internet finance, Bank of Nanjing is actively transforming its business by building its own Internet finance platform.
Li Yong
Deputy General Manager, Information Technology Department, Bank of Nanjing
“The OceanBase database system has been tested in many Internet finance scenarios within Ant Financial. This gave us the confidence to try it. Our experience has shown that by choosing the OceanBase database, Bank of Nanjing has provided a more solid foundation for the 'Xinyun+' Internet finance platform.”
Business challenges
Online horizontal scaling: The ability to quickly scale out hardware resources without service interruptions.
High-concurrency processing: The ability to handle sudden spikes in traffic, such as those that occur during the Double 11 shopping festival.
Software, hardware, and O&M costs: The need to significantly reduce costs while meeting the previously mentioned requirements.
Optimization results
On September 28, 2017, Bank of Nanjing, Alibaba Cloud, and Ant Financial signed a strategic cooperation agreement. They jointly launched the Bank of Nanjing 'Xinyun+' open platform for Internet finance. The 'Xinyun+' platform is the first major result of the collaboration between Alibaba Cloud and Ant Financial's Finance Cloud. By building the 'Xinyun+' platform, Bank of Nanjing's core Internet finance system achieved significant improvements in the following areas:
Scalability: OceanBase performed multiple online horizontal scale-outs during and after the platform's construction.
Processing capacity: Increased from under 100,000 transactions per day to over 1 million transactions per day.
Cost reduction: The maintenance cost per account dropped from CNY 30–50 to CNY 4.
MYbank
Company profile
MYbank positions itself as the preferred financial service provider for online merchants, a pioneer in Internet banking, and a practitioner of inclusive finance. It serves small and micro enterprises, individual consumers, rural businesses and farmers, and small and medium-sized financial institutions. It is also the first bank in China to build its core system on a finance cloud. Using a financial cloud computing platform and the mass storage capabilities of OceanBase, MYbank can handle high-concurrency financial transactions, massive big data, and elastic scaling. This allows MYbank to leverage the advantages of the Internet and big data to provide financial services to a larger number of small and micro enterprises.
Tang Jiacai
CTO, MYbank
“MYbank chose the OceanBase three-region, five-data-center deployment architecture. This not only upgraded our data protection from intra-city data center failure protection to cross-city disaster recovery, but its built-in multi-tenant data isolation also meets the management and operational needs of all our application systems, which makes designing an active-active application architecture exceptionally simple.”
Business challenges
Achieve city-level disaster recovery capabilities to meet regulatory requirements while minimizing the workload of deploying, operating, and maintaining the IT infrastructure for disaster recovery to reduce system O&M costs.
Provide a standard, secure, and efficient multi-tenant database isolation environment and management tools to meet the operational needs of all bank-wide application systems, including the core system for deposits, loans, and remittances.
Optimization results
By choosing the OceanBase three-region, five-data-center deployment architecture, MYbank achieved active geo-redundancy for business applications between Hangzhou and Shanghai. This greatly increased the system's overall throughput. For disaster recovery, the system provides lossless, automatic failover without manual intervention in the event of a server, data center, or city-level failure. The recovery point objective (RPO) is 0, and the recovery time objective (RTO) is under 30 seconds. This significantly reduces the workload for operating and maintaining IT infrastructure, which in turn lowers O&M costs.
OceanBase demonstrated high scalability by performing multiple online horizontal scale-outs during and after platform construction.
Using the multi-tenancy feature of OceanBase, MYbank implemented a resource allocation policy on the cluster based on business criticality and traffic ratios. This achieved an optimal balance between resource sharing and isolation, which greatly reduced IT infrastructure procurement costs. In addition, using the OceanBase Cloud Platform for O&M, daily operations and maintenance are managed entirely through a graphical user interface. This has significantly lowered maintenance and operational costs.
Alipay
Company profile
Alipay is a leading third-party payment platform in the Chinese mainland. It is dedicated to providing simple, secure, and fast payment solutions. During the 2017 Double 11 shopping festival, it processed a peak of 256,000 transactions per second. All of Alipay's core business data, including transactions, accounts, Huabei, and Jiebei, are stored on OceanBase. Compared to traditional Oracle solutions, OceanBase achieves higher scalability at a lower cost. This capability helps Alipay smoothly handle peak traffic during major sales promotions.
Cheng Li
CTO, Ant Financial
“OceanBase has provided stable support for Alipay's core transactions, payments, and accounting. It has been tested through multiple Double 11 events and has established a high availability (HA) architecture deployed across data centers and regions. It has also played a vital role in daily operations, emergency drills, and disaster recovery switchovers.”
Business challenges
Consistency: Consistency is the lifeline of financial services. To handle hardware or system failures, such as data center, OS, or machine failures, traditional databases offer several options. The maximum availability mode may cause data loss if the primary database fails. The maximum protection mode increases annual downtime and reduces performance.
Scalability: Traditional hardware-based scale-up solutions are very expensive. Ant Financial uses a sharding method with its self-developed middleware, ZDAL, to hide table sharding details and present a single-table view to the business application.
Availability: Financial services have extremely high availability requirements, typically exceeding 99.99%. Some financial institutions use the native features of their databases to provide system availability. For example, Oracle offers two HA solutions: RAC and DataGuard. However, recovery time in a failure scenario can be long. Therefore, businesses often implement their own HA solutions, such as Failover, to improve fault recovery time. This approach also introduces significant complexity.
Cost and performance: The costs for traditional databases include hardware and license fees. Unlike traditional financial enterprises, Internet finance companies serve an enormous number of users. As a result, traditional pricing models lead to very high costs.
Optimization results
OceanBase made several improvements to ensure consistency. At the architectural level, it introduced the Paxos protocol, multiple data validation mechanisms, and an improved Alipay business model. These mechanisms work together to ensure financial-grade consistency.
The HA strategy of OceanBase is very different from traditional shared-storage solutions. OceanBase uses a Share-Nothing architecture, where each component has its own continuous availability plan.
The deployment architecture is also different. Alipay's order-related services use a 'three-data-center, one-city' deployment. This provides disaster recovery for single-machine and single-data-center failures. It uses RFO (Recovery Failover Objective) for remote disaster recovery, achieving an optimal balance between performance and availability. Accounting services use a 'five-data-center, three-region' deployment. In addition to providing disaster recovery for single-machine and single-data-center failures, it also provides automatic disaster recovery for city-level failures. In both multi-zone and active geo-redundancy disaster recovery scenarios, the RPO is 0 and the RTO is under 30 seconds.
Taobao
Company profile
Alibaba is one of the world's largest e-commerce websites. During the 2017 Tmall Double 11 event, the total transaction value for the day was CNY 168.2 billion. The Favorites feature on Taobao and Tmall is very popular among users. When browsing the Taobao website, users can add their favorite products or shops to their Favorites list to find them again quickly. Users can also share their favorite products or shops with friends. Currently, the Taobao Favorites feature manages several hundred terabytes of data and serves 800 million Taobao users.
Lin Yubing
Basic Transactions, Technology Department, Taobao
“The Favorites service supports over 50 business parties within the group. It manages nearly 100 billion favorite relationships with a concurrent load of hundreds of thousands of operations. OceanBase provides excellent support for the read and write scenarios of the Favorites feature. It has passed high-concurrency tests during multiple major sales promotions. It runs stably with high throughput, excellent performance, and low cost, perfectly meeting the business development needs of the Favorites feature.”
Business challenges
The Favorites feature handles tens of millions of writes per day and must support a peak write rate of tens of thousands of transactions per second.
Queries for the Favorites feature are join queries between favorite records and product information. On average, each query joins hundreds of records. The peak display rate for users during Double 11 can reach hundreds of thousands per second, which places strict demands on database performance.
Optimization results
Using the advanced distributed features of OceanBase, data from a single table is automatically distributed across dozens of low-cost microservers. These servers work together to support high-intensity daily writes and easily handle the write pressure.
Using the excellent disaster recovery features of OceanBase, the system is deployed across three data centers. This ensures that user access is not affected, even if an entire data center fails.
Alimama
Company profile
The Alimama advertising business is primarily a Pay-for-Performance (P4P) advertising system. The report center is the only platform for Alimama to display ad performance data to advertisers, which acts as a guide for ad placement. Within the diverse commercial scenarios of the Alibaba platform, it provides high-quality, efficient, and reliable data services to customers. The report platform categorizes and summarizes a wide variety of commercial advertising information. It provides report services for business lines such as Express, Diamond Booth, Brand Performance, One-stop, Native Content, and New Single Product. It offers a variety of precise, multi-dimensional ad performance analysis services for sellers on the Alibaba business platform.
Zhang Weiyu
Director, Basic Shared Technology Development Platform, Alimama
“OceanBase has successfully met the big data processing needs of our advertising business, including requirements for storage system scalability, parallel computing, statistical computing, high throughput, low latency, and resource isolation. In the evolution of our reporting business, it has helped us build a universal system that separates business from the platform and is oriented toward performance indicator development.”
Business challenges
Developer efficiency: The report platform handles the summarization and display of a wide variety of advertising data on the Alibaba business platform. Different business lines have different reporting requirements. Even within the same business line, different marketing scenarios lead to different dimensions of data abstraction and encapsulation. As the report development process evolved, the platform gradually separated the business logic from the system. It shifted from a report-oriented development model to a universal solution oriented toward indicators. This breaks down report development into combinations of fine-grained indicators. The computing and storage models that different indicators rely on can vary greatly depending on business characteristics. The rich partitioning methods and online analytical processing (OLAP) capabilities provided by OceanBase effectively solve the problem of building business indicators in different scenarios. For our developers, this means they can focus more on what indicators they need, rather than how to retrieve the data from the storage system.
Big data processing capability: With the rapid growth of Alibaba Group's business, the importance of promotional marketing in driving commercial traffic has become more apparent. As the final link in the marketing product chain, reports have increasingly diverse and personalized requirements. The volume of report data has grown to the terabyte or even tens of terabytes scale in recent years. At this point, the scalability of the storage system becomes very important. Estimating storage resources for 5 to 10 years in advance could lead to serious resource waste in the early stages when the data scale is small. Conversely, if the initial estimate was too low, the data migration required for scaling out a MySQL and middleware cluster would be time-consuming and labor-intensive. At the same time, to provide users with a good data display experience, we require that each data calculation is completed quickly (usually in under 10s). For some big data read and write requests, it is difficult to meet this requirement without parallel computing. However, parallel queries on big data cannot slow down high-priority small requests in the system. Additionally, when the data size of a single MySQL table exceeds 20 million rows, its query performance drops sharply. This is a major drawback that the business cannot tolerate. Therefore, for our system selection, we preferred a storage solution such as OceanBase that offers high throughput, read/write isolation, and resource isolation.
Ease of use: The advertising business is a typical online analytical processing (OLAP) business. It must analyze the relationships between vast amounts of buyer and advertising data to accurately assess the effectiveness of an advertiser's ad placements. Therefore, the report platform involves many multi-dimensional data join queries and big data grouping and aggregation queries. It also involves some specialized statistical function calculations. The SQL capabilities of popular ROLAP and MOLAP analytical data query solutions in the advertising field are not user-friendly enough. Therefore, we would need to build a complex layer of business abstraction on top of their APIs and encapsulate it into a universal software development kit (SDK) for the business. This means we have to invest more developers and maintenance personnel in this cumbersome SDK, which greatly reduces development efficiency. Therefore, we also needed a storage system with robust support for the SQL language.
System cost: Another solution is to use the RAC solution provided by Oracle, which is a common choice for commercial companies. It uses shared storage to scale out data storage space and adds computing nodes on the shared storage to provide high-speed parallel processing. This solution is based on expensive hardware and Oracle database license fees. This approach did not align with our goal of building a low-cost technology stack.
Optimization results
As a universal distributed relational database system, OceanBase provides rich partitioning methods (such as HASH, RANGE, and RANGE+HASH). It also provides online, business-transparent dynamic partitioning capabilities. To scale out the cluster, a database administrator (DBA) can simply add storage nodes and perform some simple DDL operations. This process is completely transparent to the business and solves the problem of explosive data growth.
OceanBase is compatible with most features of MySQL 5.6, fully covering the needs of the reporting business. Business applications can use OceanBase just like MySQL without requiring major logical changes. As a distributed relational database, it can also provide complex distributed JOIN capabilities across multiple nodes, along with parallel aggregation, sorting, and rich mathematical function capabilities. This meets our computing needs in most scenarios. In addition, OceanBase has also customized an approximate computing feature for the report platform. For operations on extremely large result sets, OceanBase selects data points that have a greater impact on precision and performs aggregate calculations based on them. For calculations on very large datasets, this feature can quickly produce an approximate result that is very close to the exact value.
As a horizontally scalable distributed relational database system, all nodes in an OceanBase cluster are peers. Each node can provide both read and write capabilities, greatly improving the overall throughput of the system. This also meets our need to import data quickly, with a peak of over 100,000 Transactions Per Second (TPS). At the same time, each node can be deployed on a low-cost PC server. Therefore, the cost-performance ratio is dozens of times better than that of the RAC solution.