This article compares five major Chinese domestic databases across three technical generations—traditional centralized (Dameng, Kingbase), open-source modified (openGauss), and self-developed distributed (OceanBase, TiDB)—detailing their kernel origins, compatibility targets, strengths, weaknesses, and a decision framework based on compliance requirements, existing workload, scale, and operational readiness.
The article argues that manual database sharding is shifting from a default scaling strategy to a fallback option, as native distributed databases like TiDB and OceanBase internalize sharding, routing, and distributed transactions, reducing application complexity, though they introduce higher latency, operational overhead, and cost, making single-node MySQL with proper indexing preferable for smaller datasets.
The article explains why sharding should be a last‑resort scaling option, describes vertical and horizontal partitioning, compares hash, range and directory sharding schemes, discusses hotspots, routing, data colocation, migration steps and tool choices, and warns about the operational complexity involved.
MetaDB, the unified metadata layer for Baidu Cloud Storage, was accepted to NSDI'27 after more than five years of production use, delivering up to 10.7× higher same‑directory throughput, halving the required metadata machines, and addressing three fundamental shortcomings of generic distributed databases in handling file‑system metadata.
The article systematically breaks down pagination challenges in billion‑row sharded databases, compares common sharding strategies and middleware architectures, analyzes the performance drawbacks of a naïve global‑query approach, and presents several practical alternatives—including keyset pagination, two‑stage queries, index‑table tricks, and external search or NewSQL solutions—while highlighting their trade‑offs for interview discussions.
The article analyzes the write bottleneck of processing 30,000 payment orders per second on an Oracle‑Java‑Spring stack, explains why sharding was used historically, why MQ‑based peak‑shaving is discouraged, and compares self‑built sharding with modern distributed databases such as PolarDB‑X, TiDB and OceanBase, while summarizing public practices from major Chinese tech firms.
The article systematically compares TiDB, OceanBase, and PolarDB across architecture, consistency mechanisms, scalability, high‑availability, and operational characteristics, highlighting each system’s strengths, limitations, and the business scenarios they best serve, and offers a practical decision‑making framework for selecting the right distributed database.
This article explains how Volcano Engine's veDB leverages compute‑storage separation, Kubernetes operators, and declarative operations to achieve extreme deployment density, seamless scaling, and high‑availability for millions of database instances, while addressing the challenges of traditional VM‑based deployments.
This article objectively compares middleware‑based sharding solutions with native NewSQL distributed databases, examining their architectural differences, transaction handling, high‑availability, scaling, SQL support, storage engines, and maturity to help engineers decide which approach best fits their workload.
With massive data growth, sharding databases introduces pagination challenges; this article explains the root cause of pagination issues in sharded environments, presents the global query method, its pros and cons, and introduces the no‑skip optimization to improve performance while maintaining data accuracy.
This article dissects the challenges of implementing pagination on billion‑row sharded tables, explains common sharding strategies, compares SDK, Proxy, and Sidecar architectures, and presents practical solutions such as global query, infinite scroll, two‑phase queries, index tables, and external storage to help engineers ace interview questions.
The article explains why the last six digits of Taobao order numbers are constant, revealing the “gene” method that embeds a hashed user identifier to bind orders to specific shards, thereby improving query performance, ensuring balanced data distribution, and supporting scalability in large e‑commerce systems.
This article explores the origins of the "no more than three‑table JOIN" rule, explains MySQL's join algorithm limitations, shows real‑world performance pain points in distributed architectures, and presents practical solutions such as step‑wise queries, denormalization, and materialized views to safely break the restriction.
Zhihu’s evolving data architecture transitioned from a single MySQL setup to a heterogeneous ecosystem of SQL, NoSQL, and graph databases, ultimately adopting OceanBase for its multi‑tenant, high‑compression capabilities, supported by tools like OMS, OCP, and ob‑operator to achieve cost savings, performance gains, and robust operations.
At ICDE 2025 in Hong Kong, OceanBase showcased six accepted papers—including a Best Industry and Application Paper Runner‑Up award—while also hosting a technical symposium on AI‑era databases and presenting detailed abstracts on distributed database innovations, secure query processing, federated learning, hypergraph clustering, query equivalence, and cardinality‑estimation workload generation.
This article objectively compares middleware‑based sharding with NewSQL distributed databases, examining architecture, distributed transactions, performance, high availability, scaling, SQL support, storage engines, and ecosystem maturity to help architects decide which solution fits their specific workload and operational constraints.
The article objectively compares NewSQL distributed databases with middleware‑plus‑sharding solutions, covering architectural differences, distributed transaction handling, high‑availability, scaling, SQL support, storage engines, maturity, and provides a decision‑making checklist to help engineers select the most suitable approach for their workloads.
This article objectively compares NewSQL distributed databases with traditional middleware‑based sharding solutions, examining their architectures, distributed transaction support, high availability, scaling, SQL capabilities, and maturity to help readers decide which approach best fits their workload and operational constraints.
An in‑depth comparison of NewSQL databases and middleware‑based sharding reveals each approach’s architectural strengths, distributed transaction handling, scalability, HA mechanisms, storage engine design, and ecosystem maturity, guiding readers on when to adopt NewSQL versus traditional sharding solutions.
This article explains how Google Cloud Spanner combines relational database features with NoSQL scalability, using multi‑version storage, TrueTime, Paxos consensus, and dynamic sharding to deliver a globally distributed, strongly consistent, high‑availability database solution.
The article outlines the growing difficulties of managing diverse databases in cloud-native environments, introduces Apache ShardingSphere as a comprehensive open‑source solution with three core capabilities—connectivity, enhancement, and pluggability—and guides readers through a three‑step learning path from fundamentals to deployment and testing.
This article objectively compares NewSQL databases with middleware‑based sharding solutions, examining architecture, distributed transactions, CAP constraints, high availability, scaling, SQL support, storage engines, and maturity to help readers choose the right approach for their workloads.
This article objectively compares NewSQL distributed databases with traditional middleware‑based sharding solutions, examining their architectures, transaction models, scalability, high‑availability mechanisms, storage engines, maturity, and offering a decision framework to help engineers choose the most suitable approach for their workloads.
The article objectively compares NewSQL distributed databases with traditional middleware‑based sharding solutions, examining their architectural differences, distributed transaction support, performance, scalability, high‑availability mechanisms, storage engines, and practical suitability for various application scenarios.
This article explains what distributed databases are, outlines their four key characteristics—high performance, scalability, high availability, and data consistency—and reviews five prominent systems (OceanBase, TDSQL, Google Spanner, CockroachDB, and TiDB) that illustrate these concepts in real‑world applications.
The article presents LightPool, a cloud‑native storage‑pooling solution that leverages NVMe‑over‑Fabric, Kubernetes‑based scheduling, and a lightweight user‑space engine to deliver high‑performance, low‑cost, and highly available storage for large‑scale distributed databases while eliminating traditional bottlenecks.
The article introduces LightPool, an open‑source NVMe‑over‑Fabric storage pool system presented at HPCA 2024, which combines cloud‑native design, high‑performance lightweight storage engine, and advanced scheduling to improve resource efficiency, cost, and availability for large‑scale distributed databases.
The article introduces LightPool, an open‑source cloud‑native storage‑pool architecture built on NVMe‑over‑Fabric that delivers high performance, low cost, and high availability for large‑scale distributed databases, and explains its design, scheduling, storage engine, and hot‑upgrade/migration capabilities presented at the 30th IEEE HPCA conference.
This article introduces database middleware, explains why it’s needed for scaling and performance, and provides a detailed overview of six popular middleware solutions—including ShardingSphere, MyCat, MySQL‑Proxy, DRDS, Atlas, and Zebra—highlighting their core features and use cases.
This article explains the storage‑compute separation architecture introduced in StarRocks 3.0, presents industry case studies, details the design of StarOS and compute nodes, discusses technical challenges and key techniques, and evaluates cost, reliability, elasticity, and performance through benchmarks and user feedback.
This article examines the current landscape of Chinese databases, compares centralized and distributed architectures, outlines when distributed solutions are truly needed, provides performance test data, and offers practical advice on sharding, SQL design, and avoiding cross‑node bottlenecks.
Database ArchitectureShardingcentralized vs distributed
This article surveys emerging database platforms such as PlanetScale, YugaByteDB, Dolt, CockroachDB, Cloudflare D1, Xata, SurrealDB, and FaunaDB, highlighting their architectures, features, pricing, and trade‑offs for developers seeking modern, distributed, or serverless solutions.
This article examines the challenges of legacy mainframe‑based banking core systems, explains why banks are shifting to general‑purpose servers, outlines the distributed architecture and database requirements, and showcases GaiaDB‑X’s multi‑layer design and real‑world financial deployments.
This article examines the trade‑offs between centralized and distributed OLTP databases, presents industry usage statistics, performance benchmarks, practical questions for migration, and detailed guidance on sharding, SQL design, and operational considerations to help decide when a distributed solution is truly needed.
The article reviews 70 years of database evolution, outlines the two historic generations of distributed databases, highlights unresolved issues of correctness, performance and usability, and proposes a forward‑looking “third‑generation” vision that seeks 100 % data integrity, top speed, and true ease of use.
The article examines the challenges of running MySQL in Docker containers, highlighting data‑security, performance, state, and resource‑isolation concerns, while also presenting scenarios and strategies where containerizing MySQL can be viable.
This article defines distributed databases, examines their external traits such as write‑heavy, low‑latency, massive concurrency, massive storage and high reliability, explores internal architectures like client‑side sharding, proxy middleware and unit‑based designs, compares them with Amazon Aurora, and summarizes key takeaways.
The article explains that the core of the programmer's 35‑year crisis is the rapid pace of technological change outstripping learning speed, illustrated with backend sharding, frontend toolchain shifts, and AI advances, and offers practical advice to broaden skills beyond narrow tech stacks.
This article examines hybrid deployment schemes for distributed databases across X86 and ARM server architectures, detailing cross‑cluster, multi‑data‑center, and single‑data‑center approaches, and discusses their applicability to financial industry workloads, performance testing, and gradual migration strategies.
This article examines why traditional single‑node databases struggle with growing data volumes, outlines the three main distributed‑database architectures, compares their trade‑offs in availability, consistency, scalability and operational complexity, and offers practical criteria for deciding whether a distributed solution is truly needed.
This article explains the architecture and key components of distributed analytical databases, contrasting them with MPP systems, and details the design of StarRing’s ArgoDB—including its storage, compute, transaction, and SQL engines—highlighting scalability, elasticity, and resource management for modern data warehousing.
This article examines the architecture of distributed databases, identifies the challenges of global sorting across shards, and presents both in‑memory and disk‑based proxy sorting solutions, detailing buffer configuration, merge‑sort and priority‑queue techniques, performance trade‑offs, and practical constraints for large‑scale queries.
The article examines three hybrid deployment schemes—cross‑cluster, multi‑data‑center within a cluster, and single‑data‑center within a cluster—for distributed databases running on both X86 and ARM servers, highlighting their technical steps, benefits, and relevance to the financial industry.
The article explains the limitations of standalone MySQL for large‑scale workloads, explores sharding, read/write separation, and middleware approaches, and argues that modern distributed databases and cloud‑native solutions are the sustainable path forward for data‑intensive applications.
The article explains why auto‑increment primary keys are unsuitable for sharded databases, recommends globally unique ordered identifiers, and details index and global‑table strategies—including redundant tables, index tables, and embedding shard information in keys—to achieve efficient single‑shard queries.
The article traces the evolution from Codd's relational model to modern RDBMS scaling limits, explains why centralized Hadoop/HBase architectures struggle with high‑concurrency workloads, and shows how Cassandra’s decentralized design—using consistent hashing, gossip, and virtual nodes—overcomes these bottlenecks while offering flexible consistency guarantees.
Even non-programmers notice MySQL’s growing pains—large tables, oversized databases, and read-heavy workloads—prompting a shift toward distributed databases, where sharding, read/write separation, and cloud-native solutions like TiDB or Aurora replace traditional middleware, marking the end of single-node MySQL dominance.
This article presents a comprehensive overview of TDSQL‑PG's HTAP capabilities, detailing the evolution of its storage, compute, transaction management, hybrid row‑column storage, cost‑based optimizer, vectorized execution, and resource isolation strategies for handling mixed OLTP and OLAP workloads.
The article explains how to choose and implement routing keys for database sharding in an online food ordering system, covering supported scenarios, user‑side and merchant‑side keys, single‑ and multi‑database strategies, and the impact on data distribution and query efficiency.
This article explains the background, challenges, and design principles of horizontal scaling for TDSQL, detailing its architecture, scaling process, shard‑key selection, high availability, distributed transactions, performance optimization, practical case studies, and a Q&A section.
Effective index and primary key design in distributed database architectures requires using globally unique, ordered identifiers instead of auto‑increment keys, embedding shard information in keys, employing auxiliary index tables, and handling global tables to ensure queries target a single shard, thereby preserving scalability and performance.
This article explains why using auto‑increment primary keys in a distributed database leads to collisions and performance issues, and presents practical strategies such as ordered UUIDs, embedding shard keys, index tables, and global table replication to achieve efficient, scalable index design.
Index Optimizationdistributed databasesglobal UUID
This article explains the principles of Dynamo‑style leaderless replication, strict and loose quorum, version repair, and how Ctrip Hotel applied these concepts in its multi‑media storage systems Hare and InfoKeeper to achieve high availability and performance.
The article explains why auto‑increment primary keys are unsuitable for sharded databases, recommends using globally unique ordered identifiers such as UUIDs or Snowflake IDs, and discusses index design strategies—including index tables, embedding shard keys, and handling global tables—to ensure efficient single‑shard queries.
The article highlights how distributed databases are essential for the financial industry’s high‑performance, scalable, and highly available data needs, outlines a webinar covering expert selection criteria, implementation tactics, integration challenges, cloud‑native adoption, and real‑world case studies for regulated financial environments.
This article traces the rise of domestic databases in China, highlights the technical and market challenges they faced, and explains how Huawei’s GaussDB leverages cloud, distributed architecture, and ecosystem strategies to become a competitive, high‑availability solution for enterprises.
This article examines the architecture of distributed databases, identifies challenges in sorting large result sets across shards, and presents step-by-step memory‑based, disk‑based, and incremental proxy sorting techniques, including buffer sizing, priority‑queue merging, and trade‑offs to avoid OOM and excessive I/O.
In distributed databases, proxies must merge sorted results from multiple shards, but large result sets exceed memory limits; the article proposes a batch‑fetching approach using per‑shard sort buffers and a priority‑queue merge, eliminating disk I/O and reducing network waste while preserving global order.
Data PartitioningDatabase ArchitectureSorting Algorithms
The article examines the strikingly similar development trajectories of electric vehicles and distributed databases, tracing their historical origins, technological challenges, market dynamics, and recent breakthroughs such as Tesla's autonomous electric cars and OceanBase's native HTAP capabilities, while highlighting the role of open‑source and scaling strategies.
This article explains TiDB's SQL layer architecture, details how SQL requests are processed and optimized, demonstrates the use of EXPLAIN and EXPLAIN ANALYZE, and presents two real‑world case studies that illustrate common performance pitfalls and practical optimization techniques.
The article explains what an ecosystem means, then defines a distributed‑database ecosystem and uses TiDB as a case study to detail upstream/downstream migration tools, backup and recovery solutions, monitoring and alerting, daily operational utilities, the TiDB Operator for Kubernetes, automation platforms, and emerging projects built on TiDB components.
This article explains the concepts, types, advantages, and challenges of database sharding—including vertical and horizontal partitioning—covers when to consider splitting, practical strategies for handling transactions, joins, pagination, global IDs, and lists popular middleware solutions for implementing sharding in production.
Global ID GenerationHorizontal ShardingVertical Sharding
This article traces the evolution of database systems—from the first network and hierarchical models through relational databases, NoSQL sharding, and finally modern distributed SQL—explaining why distributed databases emerged, how they handle data distribution, consistency, SQL-to‑KV mapping, and transaction challenges.
The article explains how distributed database systems appear as a single logical database but operate over a network of nodes, detailing failures, stale reads, consistency models, architecture anomalies, sharding, replication, and various read/write scenarios that illustrate the complexity of achieving strong consistency.
NetBank’s database deployment has evolved from a simple two‑site three‑center disaster‑recovery model to a sophisticated three‑site five‑center architecture, incorporating distributed databases, multi‑tenant isolation, transaction consistency, latency optimization, and containerized deployment to achieve high availability, scalability, and cost efficiency.
This article details NetBank's evolution of database deployment—from early distributed setups to a unitized, cloud‑native architecture—covering disaster‑recovery upgrades, distributed database design, multi‑tenant strategies, containerized migration, and the performance and operational impacts of moving to a three‑site five‑center model.
This article provides a detailed summary of database concepts including data models (relational, document, graph), storage engine architectures (page‑oriented B‑tree, log‑structured LSM), transaction mechanisms, isolation levels, distributed transaction protocols, partitioning strategies, indexing, consistency models, and consensus algorithms such as Lamport timestamps and Raft.
This article examines the evolving data‑storage demands of large‑scale commercial advertising systems, traces the development of BaikalDB from early MySQL sharding to a heterogeneous, cloud‑native distributed database, and explains its storage, compute, and scheduling designs that deliver high reliability, low cost, and millisecond‑level performance.
This article explains why code review is critical, outlines its benefits and common difficulties, describes the TDSQL‑C distributed database architecture, and presents a detailed code‑review workflow, author and reviewer responsibilities, commit conventions, and cultural practices to improve software quality and reliability.
This article introduces the fundamentals of distributed database architectures, compares Shared‑Nothing and Shared‑Storage designs, explains their three‑tier structure, core engine components, SQL execution flow, performance and cost optimizations, and showcases a real‑world high‑traffic deployment in Douyin’s Spring Festival event.
This article explains the differences between relational and NoSQL databases, outlines the definition, benefits and characteristics of distributed databases, and reviews ten widely used distributed database products such as TiDB, Google Spanner, CockroachDB, YugabyteDB, OceanBase, and others.
The Volcano Engine Developer Community’s May 15 online Meetup showcased four technical sessions—Service Mesh traffic management, distributed databases, mobile file distribution, and IaaS virtualization—detailing how Douyin’s Spring Festival red‑packet event handled massive traffic and ensured high performance.
The talk traces the evolution of distributed databases—from early shared‑memory and shared‑nothing designs through New‑SQL breakthroughs—to Tencent’s HTAP‑oriented TDSQL, highlighting its strong‑consistency architecture, anomaly taxonomy, and performance advantages over systems like Google Spanner while addressing modern scalability and hardware challenges.
In a May 15 meetup, ByteDance engineer Ma Haoxiang discussed his background, the culture at ByteDance, recommended resources, and detailed how distributed databases differ from traditional relational databases, highlighting their massive capacity, low cost, high performance, and the specific performance and disaster‑recovery challenges faced during Douyin’s Spring Festival red‑envelope activity.
This article objectively compares NewSQL distributed databases with traditional middleware‑plus‑sharding solutions, examining architecture, distributed transactions, CAP constraints, high‑availability, scaling, storage engines, maturity, and practical adoption scenarios to help readers decide which approach best fits their needs.
This article objectively compares NewSQL distributed databases with middleware‑based sharding solutions, examining their architectural differences, distributed transaction handling, high‑availability mechanisms, scaling and sharding strategies, SQL support, storage engines, and overall maturity to help readers decide which approach best fits their workload.
The article reviews the evolution of database systems driven by cloud computing, big‑data demands and distributed architectures, highlights Alibaba Cloud’s cloud‑native offerings such as PolarDB and AnalyticDB, and discusses trends, security, and best practices for modern enterprise data platforms.
This article introduces Apache ShardingSphere's architecture—including ShardingSphere‑JDBC, ShardingSphere‑Proxy, and the upcoming Sidecar—explains data sharding principles, read/write splitting, and provides a complete Spring Boot example with Maven dependencies, Java source code, and configuration files for horizontal database partitioning.
The article introduces the JUST engine, a spatio‑temporal data platform that extends GeoMesa with three new indexes (Z2T, XZ2T, time_range), defines nine common and three specialized data models, provides default indexing strategies, and offers detailed SQL usage guidelines for efficient querying of massive urban datasets.
This article explains the principles of hash tables, how hash functions are used for data sharding in distributed databases such as DBLE, compares hash‑modulo and consistent‑hash approaches, and discusses their advantages, limitations, and practical configuration tips.
This article recounts the 2007 challenges faced by Taobao's monolithic MySQL setup and explains how the TDDL middleware introduced SQL optimization, sharding, read‑write separation, caching, and dynamic scaling to transform the system into a distributed, high‑performance database solution.
This article explains the CAP theorem, the trade‑offs among consistency, availability, and partition tolerance, introduces the BASE model and its properties, and details how different database systems implement various consistency guarantees such as strong, eventual, and causal consistency.
This article explains MySQL replication (master‑slave and master‑master), data sharding principles and implementations, and various distributed database deployment architectures, covering replication mechanisms, advantages, failure recovery, sharding strategies, middleware like Mycat, and scaling solutions for high‑availability systems.
In a June 10 live session, ICBC's software development center manager Wei Yadong presented the financial industry's distributed database needs and selection criteria, answering 31 detailed questions covering architecture, middleware, sharding, high availability, migration, and operational practices.
Apache ShardingSphere, the open‑source distributed database middleware originally contributed by Dangdang and now backed by the Apache Software Foundation, has graduated to a top‑level Apache project, showcasing its evolution from a simple JDBC sharding driver to a comprehensive ecosystem with JDBC, Proxy, and Sidecar components, adopted by over 120 companies.
This article objectively compares NewSQL databases with middleware‑based sharding solutions, examining architecture, distributed transactions, CAP constraints, performance, high availability, scaling, SQL support, storage engines, maturity, and provides practical guidance for choosing the right approach.
This article examines the pressures driving banks to redesign their core database systems, compares three distributed‑access approaches, explains the components of modern distributed databases, analyzes two‑phase and three‑phase commit issues, and evaluates consensus algorithms, CAP/BASE trade‑offs, GTM design, and point‑in‑time recovery.
This article reviews the evolution, market trends, core components, architectural challenges, and emerging technologies of cloud‑native distributed database systems, highlighting Alibaba Cloud's solutions such as POLARDB, AnalyticDB, and AI‑driven management platforms that enable elastic, high‑availability, and intelligent data services for modern enterprises.
This article explains the primary use cases for distributed databases, contrasts them with traditional databases, and describes how data partitioning and metadata enable clients to locate data without scanning all nodes, highlighting both external user benefits and internal implementation challenges.
Data PartitioningHigh Availabilitydistributed databases
This article presents a comprehensive overview of the technical architecture behind a commercial advertising platform, covering business fundamentals, microservice migration, service governance, distributed storage, real‑time billing, and large‑scale reporting, and explains the practical engineering choices made at Alibaba.
This article explains linearizability in distributed systems, compares Raft read strategies, and describes how CB‑SQL uses range leases and node leases to achieve strong consistency while reducing overhead, illustrating the trade‑offs and evolutionary design choices behind its consensus mechanisms.
This article details Industrial Bank's multi‑year IT architecture transformation, describing the challenges of legacy mainframe‑based OLTP, the strategic shift to a distributed MySQL ecosystem, the implementation phases, high‑availability designs, containerization efforts, measurable outcomes, and future directions for cloud‑native and data‑exchange capabilities.
The article explains how distributed databases can perform online, asynchronous schema changes without blocking reads or writes by using lease‑based coordination, intermediate delete‑only and write‑only states, and a multi‑step algorithm inspired by Google F1 and implemented in CB‑SQL.
This article objectively compares NewSQL distributed databases with middleware‑based sharding solutions, examining their architectures, distributed transaction handling, scalability, performance, high‑availability, and operational considerations, and provides guidance on selecting the appropriate approach based on workload, consistency, and organizational constraints.
The article outlines eight practical techniques—including caching, page staticization, database optimization, hot‑data separation, operation merging, read‑write splitting, distributed databases, and the use of NoSQL and Hadoop—to efficiently store and serve massive data volumes in large‑scale internet services.
This article explains vertical and horizontal database sharding, compares hash-modulo and range-based partitioning, and proposes a hybrid design that balances load, prevents hotspots, and eliminates costly data migrations during scaling.
The article explores TiDB’s evolution philosophy, detailing its core beliefs, early user stories, and the three‑step "Make it work, make it right, make it fast" approach, while covering technical choices such as cloud‑first design, hardware‑agnosticism, formal verification, massive testing, performance tuning, and cost‑effective scaling.
This article recounts the author’s year‑long experience transforming Sharding‑JDBC into the ShardingSphere ecosystem, detailing technical challenges, community building, Apache Incubator admission, and the project’s evolution into a financial‑grade distributed database solution.
The article details the design and overall architecture of ZhaiZhai's instant messaging (IM) system, covering product positioning, a four‑layer architecture (user, entry, logic, storage), and analyses of scalability, high availability, reliability, extensibility, and performance, highlighting the use of C++, RPC, MySQL, TiDB, and custom middleware.
This article traces the evolution of Alibaba’s distributed data layer—from the early Cobar system to the modern TDDL framework and DRDS service—explaining their architectures, limitations, sharding principles, transaction‑boundary strategies, heterogeneous index tables, and the Jingwei data‑replication platform that together enable seamless scaling and high‑performance SQL processing across thousands of databases.