Riding on the momentum, OceanBase drives digital payments lean growth

According to a related report from McKinsey, the global payment industry revenue in 2020 fell for the first time in 11 years. The black swan event has brought about a major shift in payment behavior, with a sharp drop in cash usage and the rapid transfer of business services from offline to online. These changes have brought new opportunities and challenges to payment participants.

In 2021, the payment industry revenue indicators will quickly recover and rebound. Among them, the agility of digital payment in the face of the crisis has become one of the important factors driving the faster economic recovery. For example, in Q1 of 2022, the number of registered users of Philippine e-wallet GCash exceeded 60 million for the first time, reaching the milestone of covering 83% of the adult population in the Philippines.

In this round of rapid repair and explosive growth, the battle for competitive advantage among enterprises has also begun. The entire payment industry has undergone a second transformation, from a multi-scenario single service to a full-scenario digital operation, and digital payment has changed from a convenient choice to a basic service, injecting new vitality into the digital economy and the industrial Internet.

Some enterprises have improved their cost control capabilities, improved service quality, optimized unit cost production efficiency through technological innovation, and continued to expand their own advantages and market scale through infrastructure upgrades, technical architecture iterations and the introduction of new technologies.

About the author

Author: Zhou Guiqing

OceanBase solution architect, distributed database evangelist. With nearly ten years of development experience based on traditional Oracle, he has led the BOSS core system technology architecture in the telecommunications industry, participated in the formulation of the IETF RFC8543 standard, translated and published the book "ZooKeeper, Distributed Process Coordination" by O'Reily, and has in-depth knowledge of distributed consistency protocols. Research; Worked at CNNIC, as the backbone of "National Engineering Laboratory for Internet Domain Name Management Technology", promoting the distributed transformation of the national domain name core system; former JD Cloud IaaS architect, leading the design and implementation of multiple industry solutions. Fans and enthusiasts of distributed technology.

Digital payments face new challenges

After years of development, the technical architecture of digital payment has gradually developed from a monolithic architecture to a distributed architecture such as microservices and unitization. However, few companies use native distributed databases, which lack strong business development capabilities on the core data base. Strong support.

Most of the traditional database architectures of early digital payments used stand-alone databases, because of their simplicity and ease of use, and strong ACID (Atomic atomicity, Consistency consistency, Isolation isolation, Durability persistence), helping payment companies to quickly accumulate a certain market size.

As the business grows, the capacity and performance of a single-machine database can easily reach the upper limit, and the application architecture begins to be upgraded to a distributed architecture. At the same time, the sub-database sub-table scheme based on middleware came into being. This method effectively solves the problems of data surge and high concurrency in the short term, but the essence is the horizontal combination of stand-alone databases, which also brings a series of other problems. The problem restricts the development of enterprises in the new round of growth period.

Difficulty guaranteeing business continuity

Ensuring business continuity is the top priority of digital payments, mainly including high availability and high disaster tolerance. High availability requires that the database can still ensure zero data loss and normal business operation in extreme cases such as power failure in the computer room, network abnormality, disk silence, and even fiber disconnection. High disaster tolerance refers to earthquakes, fires, and floods. When the local database encounters a disaster, the disaster preparedness database can quickly undertake the task of the local database.

When a traditional stand-alone database encounters a failure, the business may be interrupted for a long time, and even data may be lost; even if high availability and high disaster tolerance can be achieved, it depends on the integration and grading of multiple components such as the operating system, storage hardware, and database. The degree of cooperation of its own applications is low, and the switching requirements are high and difficult, which is not enough to ensure the business continuity of digital payment enterprises.

Frequent sub-database sub-table is inefficient

In the early stage of digital payment business, when the amount of data is not large, the single-machine database architecture can often run well. However, when the business develops to a certain scale, with the surge of data volume, either choose to continuously increase the stand-alone database; , a slight unreasonable design may greatly reduce the overall performance and availability of the system.

In addition, sub-database and sub-table also means that the business must be stopped within a certain period of time, which is a lot of tedious work for the operation and maintenance personnel to clean up the data, disassemble the database and disassemble the table. A large number of database instances will also lead to a linear increase in the complexity of operation and maintenance, resulting in low overall operation and maintenance efficiency.

Storage costs are getting higher and higher

Undoubtedly, with years of development and accumulation, resulting in a surge in data volume, the database cost of digital payment companies will become higher and higher.

• One is the hot data storage period required by the regulatory requirements, and the full data storage requirements for regulatory review;
• Second, business growth leads to increasing database hardware costs;
• Third, the resource utilization rate of multiple instances of sub-database and sub-table is low, resulting in waste of computing resources;
• Fourth, the accumulation of massive data storage leads to an increase in storage costs.

Therefore, the storage cost of digital payment companies is generally higher than that of other companies with lower requirements for data retention time.

Poor stability at high concurrency

With the advent of the mobile Internet, digital payment companies are required to provide stable support behind a large number of high-concurrency scenarios, such as live broadcasts and rush purchases, ultra-low discounts, and the need to call third-party payment interfaces at the same time to ensure user transactions and payments. Such behaviors are normal; for example, aggregated payments that integrate various payments such as banks and third-party payments through technical means are faced with a large number of high-concurrency scenarios every day.

The traditional database architecture needs to plan capacity in advance according to business development, does not support dynamic expansion and contraction, and does not have the capacity to carry burst traffic. In the face of a large number of SQL requests such as payments and transactions in a short period of time, if a delay of 1 millisecond is added to each SQL, the user will have a poor experience of delayed waiting, and the database is likely to directly crash.

Big data analysis process is long

China's digital payment has always been leading the global payment market in terms of market size and growth rate, and it is also based on continuous business innovation, including the transformation of scenario-based digital operations. Whether it is third-party payment or aggregated payment, there are scenarios ranging from big data analysis for B-end merchants, such as merchants querying summary data of recent transactions, statistical analysis of profit sharing data, etc.; to C-end user portraits for precision marketing scenarios, Such as payment receipt and bank data linkage.

Because the traditional architecture usually divides OLTP and OLAP into two sets of databases, when encountering scenarios that require big data analysis, it is often necessary to synchronize the data responsible for the OLTP database to the OLAP database, and then use the OLAP database for big data analysis, most of which are statistical analysis. The timeliness is T+1, which affects the timeliness of business decisions and also restricts the innovation and development of business models.

Digital payment urgently needs to upgrade the underlying infrastructure

Digital payment is facing a new round of opportunities and challenges. Facing the rising data costs, it is urgent to upgrade the most fundamental underlying infrastructure of the database.

From a stand-alone database to a middleware-based sub-database and sub-table database, the problem has not been fundamentally solved. With the explosive growth of data, not only the stability cannot be effectively guaranteed, but the performance degradation also directly brings about a decline in service experience. At the same time, the cost of database-related hardware and storage will become higher and higher, but the efficiency of computing, operation and maintenance will not be the same. There is no corresponding increase. Some payment companies have introduced native distributed databases to achieve high availability, high expansion, high performance and other architectural upgrades of the data base.

In addition to replacing the expensive stand-alone database with a lower-cost ordinary PC server, the native distributed database is fundamentally different from the traditional database at the bottom, which can effectively solve the above challenges encountered by digital payment companies.

First, at the beginning of the design of the native distributed database, it is assumed that the hardware is unreliable, and the data will be automatically scattered by the system and stored in different copies. These copies can be stored in different cabinets, computer rooms, and regions, realizing urbanization. level of lossless disaster recovery requirements. The distributed consistency protocol ensures global data consistency between multiple copies, and other copies are guaranteed to be successfully submitted after data modification, so they have high disaster tolerance capabilities.

In terms of high availability, native distributed databases also have their core advantages. For example, the native distributed database OceanBase, based on the distributed election protocol, can conduct autonomous elections when failures occur. When a minority node is down, it can switch quickly and automatically without damage, reaching RTO=0, RTO<30 seconds, reaching the highest defined by the national standard. Level disaster tolerance standard; and built-in multiple strong verification mechanisms, which can automatically detect inconsistency of multi-copy data, network data errors, disk silence errors, etc., to ensure strong data consistency.

Second, the native distributed database can achieve unlimited horizontal expansion on ordinary PC servers, and realize the linear expansion capability of computing and storage resources. Digital payment enterprises no longer need to carry out tedious manual expansion and complex sub-database and sub-table construction, which effectively reduces the operation cost. Reduce complexity and improve operation and maintenance efficiency. For example, a single table of the native distributed database OceanBase can be expanded to PB level and more than 100 billion records. Digital payment companies can configure computing resources as needed, and flexibly expand and shrink resources.

On the basis of infinite horizontal expansion, the native distributed database solves the problems of global consistency of sub-database and sub-tables, cross-database transactions, complex relational queries, and load balancing, and realizes super database computing clusters. In the case, a single cluster exceeds 1000 nodes, and the entire cluster uses MVCC and global indexes to ensure global consistency and high performance.

Third, from the expensive stand-alone database to the ordinary PC server, the hardware cost has been greatly reduced. In addition, the native distributed database has also improved the utilization rate of storage space and computing resources. For example, the original distributed database OceanBase, because of its completely self-developed storage engine, can effectively utilize the storage compression technology based on variable length and fixed length of data, the storage compression technology based on data encoding, and the low-cost storage technology based on data log separation.

Compared with MySQL and Oracle databases, the data storage volume of the same business, OceanBase is about 1/4 to 1/3 of the latter, which can effectively reduce the storage cost by 70% to 90%.

Fourth, the elastic scaling capability of distributed databases can help digital payment companies meet high concurrency smoothly. For example, the native distributed database OceanBase, based on the technology of threads and coroutines, has a very high number of connections. The number of single-point connections is 5 to 8 times that of other databases. The number of cluster connections can grow linearly according to the number of nodes.

In addition, OceanBase has the ability to release row locks in advance (ELR), allowing businesses to optimize distributed transaction response capabilities and improve the overall throughput of the system in the face of high-concurrency scenarios such as big promotions and spikes. Under the same specifications, MySQL hot row update 3000TPS, OceanBase can reach 3500~10000TPS through ELR.

Fifth, HTAP is an emerging trend in the database field in recent years, aiming to break the barriers of the original OLTP and OLAP, and make transaction processing and decision analysis more efficient. Compared with traditional databases, which can only perform OLTP and OLAP separately, distributed databases naturally have the ability to combine the two, which greatly shortens the time for big data analysis of digital payment companies.

For example, OceanBase, a native distributed database, can deploy OLTP and OLAP in the same database cluster, avoid information islands through data integration, and truly achieve a piece of data that can be processed and analyzed in real time, providing digital payment companies with faster, More accurate data analysis.

After the distributed upgrade, the effect of cost reduction and efficiency improvement is remarkable

1. Lichu Business Services: Storage costs have dropped by 74%, and profit sharing statistics time has been shortened by nearly 1 times

Lichu Business Services is one of the leading companies in aggregated payment. It owns three major product systems, namely “Sweep” in aggregated payment platform, “Fujia” in merchant light SaaS, and “Youyu” in global marketing platform. Among them, "Sweep" ranks the top two in the aggregate payment industry scale. As an innovative Internet enterprise, Lichu Commercial Service pays close attention to the development trend of emerging technologies and actively invests in practice. In order to stably support the continuous growth of future business, it chooses to carry out distributed upgrades.

The database of aggregated payment companies has to support a large number of transaction scenarios and payment scenarios every day. The accumulation of data has led to the continuous increase of the storage cost of Lichu Business Services. After upgrading to OceanBase, a distributed database, based on LSM-Tree storage architecture and adaptive advanced compression technology, after a business of Lichu Commercial Service migrated from MySQL to OceanBase, the overall storage cost dropped by 74%.

The daily order volume of Lichu Business Services is very large. For example, for agents and other scenarios, it involves summary settlement and profit sharing analysis. Therefore, there are also many needs for real-time statistical analysis while conducting online transactions. Taking the statistical analysis of profit distribution as an example, before the application of OceanBase, it usually takes 30 to 40 minutes to complete the analysis of profit distribution. This is a huge increase.

At the same time, some T+1 statistical analysis relies on other column storage databases. After OceanBase is applied, it relies on the HTAP engine to realize the ability of quasi-real-time data warehouse analysis, and also simplifies the complexity of the architecture.

2. GCash: The operation and maintenance efficiency has been greatly improved, and the data compression ratio is nearly 90%

GCash is a micro-payment system under Globe Telecom, a Philippine telecommunications company. As of Q1 2022, it has over 60 million registered users, over 5 million merchants, and over 170,000 deposit and withdrawal sites. It has become the No. 1 in the Philippines. A big e-wallet. With the explosive growth of business, GCash's MySQL cluster has reached the storage limit, requiring frequent manual splitting and cleaning of data; a large number of database instances have caused DBAs to fall into an inefficient cycle and a series of problems. Therefore, GCash decided to upgrade to a new distributed database architecture.

Through the migration service provided by OceanBase, GCash implements the complete migration of the whole-site business to the distributed database without downtime, and adopts the high-availability architecture of three availability zones to ensure the uninterrupted business of any availability zone level failure.

GCash's original hundreds of MySQL instances need to be dealt with separately when there is a problem, and the operation and maintenance cost is high. With the multi-tenant capability of OceanBase, after resource pooling, only more than 10 clusters are used to accommodate them. In addition, OceanBase provides operation and maintenance management tools (OCP), including full life cycle management of database components and related resources, monitoring and alarming, performance diagnosis, fault recovery, backup recovery, etc., which greatly improves operation and maintenance efficiency and DBA happiness.

The data of the original single database of GCash is nearly 5T. After migration to OceanBase, the data is compressed to less than 500G, and the data compression ratio is nearly 90%. The incremental data migration also has a basically similar data compression ratio, so the data storage space is significantly improved. At the same time, this also means that operation and maintenance students do not need to perform data archiving and data cleaning for a long time, and can focus on more important things.

Epilogue

Digital payment will drive the upgrade of the digital industrial model. Through the coverage of multiple scenarios, it will promote the improvement of the synergy of industrial Internet scenarios, penetrate into the supply chain and financial chain of merchants and enterprises, and open up the capital flow, information flow and logistics of merchants and enterprises. The transformation and upgrading of the business model of merchants and enterprises.

Facing a new round of opportunities and challenges, digital payment not only needs to keep pace with the times in business models, but also needs to update and iterate on infrastructure and technical architecture, eliminate outdated or middle-stage technologies, and introduce diversified technologies including artificial intelligence. , native distributed and other new technologies, reduce operation and maintenance costs, improve production efficiency, fully consolidate infrastructure capabilities, and continuously consolidate and deepen the leading edge of digital payment.