Over ten years of BAT architecture experience, shared generously!
This article compares four major distributed file storage systems—HDFS for big data batch processing, Ceph for unified cloud storage, GlusterFS for scalable file sharing, and FastDFS for high-throughput internet file serving—detailing their architectures, use cases, and suitability for different workloads.
This article explains why GPUs are essential for AI large models, detailing their parallel computing architecture, high-bandwidth memory (HBM), and multi-GPU interconnect technologies, then compares mainstream AI chips including NVIDIA's H100, H200, B200, AMD's MI300X, MI350X, and Google TPU across architecture, memory, bandwidth, and target workloads.
The article explains why idempotency is critical for Alipay’s high‑concurrency, high‑reliability payment platform and outlines four key reasons—funds safety, handling network uncertainty, user experience, and distributed architecture—followed by concrete solutions such as unique business identifiers, database unique constraints, state‑machine control, and idempotent API design.
The article explains Nginx's caching mechanism, showing how storing backend responses in memory or disk and designing proper cache keys can dramatically reduce backend load and latency, delivering performance improvements of several times to over tenfold.
The article explains that achieving stable performance for Spring Cloud Gateway at a million‑request scale requires a reactive non‑blocking model, comprehensive rate‑limiting and circuit‑breaker mechanisms, and optimized connection management and routing to handle high concurrency efficiently.
The article explains how Nginx’s event‑driven, asynchronous non‑blocking model, master‑worker multi‑process design, efficient I/O multiplexing (epoll/kqueue), and lightweight resource‑control architecture enable it to sustain extremely high concurrency with minimal overhead.
The article reviews four major distributed storage systems—HDFS, Ceph, GlusterFS, and FastDFS—detailing their architectures, core strengths such as HDFS’s batch processing, Ceph’s unified object/block/file capabilities, GlusterFS’s horizontal scalability, and FastDFS’s lightweight handling of small files, while also noting each solution’s limitations.
The article walks through a step‑by‑step Nginx performance tuning guide, covering worker_processes and worker_connections settings, Linux file‑descriptor limits, sysctl tweaks such as net.core.somaxconn and tcp_max_syn_backlog, explicit listen backlog, and zero‑copy sendfile with tcp_nopush/tcp_nodelay to dramatically increase throughput and lower CPU usage.
Alibaba’s same‑city active‑active architecture deploys multiple data centers within one city, splits traffic between them, and ensures instant failover with layered high‑availability across access, application, and data layers, using technologies like DNS, SLB, Nginx, Redis and OceanBase.
MySQL’s concurrency capacity isn’t a single number; it’s measured by QPS and TPS and varies with query complexity, indexing, transaction design, and hardware, with simple indexed reads reaching several thousand QPS on a well‑tuned server, while mixed or complex workloads often cap around one to two thousand QPS.
