This article provides a detailed technical comparison of LVS, Nginx, and HAProxy, explaining their layer operation, performance characteristics, configuration complexity, and ideal use‑cases to help architects select the most suitable load‑balancing solution for high‑traffic environments.
This article explains the roles of load balancers and API gateways, compares their layer focus, features, and performance characteristics, and outlines when to use each technology—especially in micro‑service architectures and high‑throughput network entry points.
This article examines the challenges of logging in distributed services, compares the traditional ELK stack with Baidu's Tianyan solution, details Tianyan's architecture—including Ingest, Store, Consumer, Elastic Agent, Fleet, APM, Beats, and Disruptor‑based high‑throughput pipelines—covers resource isolation, dynamic cleanup, and best‑practice recommendations for building a scalable, low‑latency log platform.
Netflix’s cloud‑gateway team redesigned its load‑balancing stack—combining client latency, server utilization, and probabilistic choice‑of‑2 algorithms—to dramatically lower error rates, improve request distribution, and enhance fault‑tolerance for millions of requests per second.
This article examines the challenges of scaling log storage and retrieval for high‑traffic services, analyzes the cost and performance limits of traditional ELK‑based pipelines, and presents a streamlined, UDP‑driven architecture using ClickHouse that dramatically reduces hardware expenses while handling hundreds of gigabytes per second.
This article provides a comprehensive overview of load balancing technologies—including LVS, Nginx, and HAProxy—detailing their architectures, operating modes, advantages, and drawbacks, and explains how to choose and configure them for high‑performance, scalable backend systems.
Replit engineers discovered that slow Docker container shutdown on preemptible VMs caused REPL sessions to hang for up to a minute, and by bypassing Docker’s kill process and directly terminating container PIDs they reduced error rates from 3% to under 0.5% and cut 99th‑percentile startup time from two minutes to fifteen seconds.
Replit engineers analyzed why container shutdown on preemptible VMs caused REPL sessions to stall for up to a minute, identified Docker's network‑release bottleneck, and implemented a direct SIGKILL workaround that reduced error rates and startup latency dramatically.
This article provides a comprehensive overview of ZooKeeper, covering its purpose as a distributed coordination service, design goals, hierarchical data model, session handling, watch mechanism, consistency guarantees, server roles, leader election, workflow of leaders and followers, and the Zab protocol that ensures reliable state replication.
This article explains how modern web services use server clusters and load‑balancing devices, compares the three most popular software balancers—LVS, Nginx, and HAProxy—by describing their architectures, operating modes, advantages, disadvantages, and suitable deployment scenarios for different site scales.
This article explains how modern web services rely on server clusters and a front‑end load balancer, compares the three most popular software balancers—LVS, Nginx, and HAProxy—by describing their architecture, operating modes, advantages, and drawbacks, and offers guidance on selecting the appropriate solution for different traffic scales.
This article details Xiaomi's journey from Kafka 0.8 to the home‑grown Talos system, covering business motivations, storage‑compute separation architecture, key challenges such as tail‑read and consistency, and extensive performance, resource, and platform optimizations that enable a high‑throughput, multi‑tenant messaging service.
This article examines Google's high‑availability distributed cron design, distills its core requirements and algorithms, and then presents a streamlined implementation for a startup using etcd and Raft, followed by a thoughtful discussion on whether early‑stage companies should adopt a middle‑platform strategy.
Levin, Didi’s open‑source GitHub project, uses shared‑memory containers and offline‑compiled binary layouts to mmap large, low‑frequency static datasets, cutting service cold‑start times from minutes to seconds, lowering memory overhead, simplifying version switches, and improving stability for high‑scale applications.
To cope with Meituan Hotel’s exploding micro‑service complexity, the infrastructure team built the Satellite System—combining MTrace and a selective, zero‑intrusion Log4j2‑based logging pipeline that streams enriched logs through Kafka, Storm, Redis and Elasticsearch, delivering second‑level trace‑log queries and six‑month retention, dramatically speeding up debugging.
This article explains the fundamentals of load balancing, compares four‑layer and seven‑layer approaches, outlines their roles and advantages, and reviews popular software solutions and typical application scenarios for modern large‑scale web infrastructures.
Uber’s CTO Thuan Pham explains how the company builds a specialized engineering team and core infrastructure in China, uses data‑center support, dynamic pricing, anti‑fraud machine learning, and safety features to adapt its global platform to local market needs.
This article summarizes Yao Yu's "Scaling Redis at Twitter" talk, detailing why Twitter chose Redis, the massive memory and QPS requirements, custom data models, Hybrid List and BTree extensions, cluster management, and operational lessons for building a high‑performance caching service.
This article examines how load‑balancing technologies such as Nginx, LVS, and HAProxy are chosen according to website scale, outlines their respective strengths and weaknesses, and describes typical deployment stages from simple single‑point setups to complex multi‑layer architectures.
