The HiLight approach inserts lightweight highlight tags into full-length inputs, training a small Emphasis Actor to score token importance and guide a frozen large language model, improving performance on tasks like recommendation and QA without modifying the solver, while keeping low latency and training cost.
The article dissects OpenAI's engineering approach to delivering low‑latency voice AI at scale, explaining why WebRTC was chosen, how a Relay + Transceiver split solves Kubernetes integration challenges, the use of ICE ufrag for deterministic routing, and how global relay and implementation choices reduce perceived latency.
The NES framework introduced by Ant Group’s CodeFuse team learns developers’ historical editing patterns to provide instant, instruction‑free edit suggestions, achieving sub‑250 ms latency and improving the flow of frequent refactoring tasks through a simple Tab‑press interaction.
This article examines how Taobao Flash Sale’s AI Agent uses Alibaba Cloud’s Tair as a high‑performance short‑term memory layer, detailing data model design, latency impact, concurrency control, elastic scaling, bandwidth handling, and TTL‑based cleanup to achieve sub‑second response times during massive traffic spikes.
This article breaks down the true bottlenecks of low‑latency WebRTC systems—network models, congestion control, memory layout, and concurrency scheduling—by examining the protocol stack, Go runtime, ICE state machine, DTLS/SRTP security, RTP/RTCP feedback, and practical high‑concurrency tuning strategies.
This article explains the fundamentals of Remote Direct Memory Access (RDMA), its low‑latency, zero‑copy and kernel‑bypass mechanisms, programming interfaces, and real‑world applications in data‑center networks, high‑performance computing, and distributed storage, providing developers with practical guidance and code examples.
In high‑frequency trading and real‑time systems, Go’s garbage‑collector pauses and channel lock contention can inflate P99 latency to dozens of milliseconds, while a disciplined Rust rewrite eliminates GC, reduces lock overhead, and achieves sub‑100‑microsecond latency with far lower CPU usage.
Qwen3‑TTS, an open‑source text‑to‑speech model from Alibaba, offers sub‑100 ms first‑packet latency, supports voice cloning, natural‑language voice design, and ten languages, can be deployed locally on a GPU with as little as 8 GB VRAM, and integrates with ComfyUI for visual workflow building.
This article explains how Apache Ignite's memory‑first architecture, distributed compute grid, and event‑driven streaming enable sub‑100 ms decision making, high throughput, and strong consistency for real‑time bidding platforms, with practical code examples, Spring Boot integration, monitoring tips, and security considerations.
This guide explains how to systematically analyze and optimize Java garbage‑collection pauses—cutting typical 200 ms stalls down to around 20 ms—by enabling detailed logs, selecting the right collector, tuning heap and generation settings, minimizing allocation, handling large objects, and balancing GC threads with CPU resources.
CosyVoice 2.0, Alibaba DAMO Academy's next‑gen high‑fidelity speech synthesis model, introduces architecture decoupling, streaming generation, reference‑audio caching and dynamic load balancing to dramatically reduce first‑packet latency and improve real‑time factor while supporting multi‑language voice cloning.
This article outlines ten practical techniques—ranging from reducing network hops and caching hot data to optimizing database queries, batching requests, trimming payloads, focusing on critical paths, and proactive scaling—to dramatically lower response times and make applications feel instantly responsive for users.
This article examines Broadcom's Scale Up Ethernet (SUE) framework, detailing how it addresses AI/HPC rack‑scale interconnect challenges by delivering ultra‑high bandwidth, microsecond‑level latency, memory‑semantic operations, and seamless compatibility with existing Ethernet infrastructure for large XPU clusters.
MyPerf4J is a high‑throughput, low‑latency Java performance monitoring tool that uses a non‑intrusive JavaAgent to collect real‑time method, memory, GC, and class metrics, offering developers quick bottleneck detection in development and continuous observability in production.
The article examines how the rapid growth of AI models and workloads is reshaping network design, highlighting the need for ultra‑high bandwidth, sub‑millisecond latency, reliability, scalable topologies like Fat‑Tree and Dragonfly, and robust security and QoS mechanisms across data‑center, cloud, and edge environments.
This article examines how to achieve low‑latency network packet processing on NVIDIA GPUs by comparing CPU and GPU implementations, exploring memory optimizations, batch strategies, stream concurrency, persistent kernels, and CUDA graphs, and presenting detailed performance measurements for each technique.
LongCat-Flash, an open‑source Mixture‑of‑Experts model released by Meituan, leverages model‑system co‑design, PD‑disaggregation, SBO scheduling and large‑scale expert parallelism within the SGLang framework to deliver dramatically lower latency, higher throughput and cost‑effective inference for AI agents, with detailed deployment instructions provided.
This article explains the limitations of TCP and UDP for real‑time applications, introduces the KCP protocol as a UDP‑based solution that combines reliability with low latency, outlines its features, working principles, code usage, and real‑world deployment examples in gaming, video conferencing, live streaming, and IoT.
Spine‑Leaf is a two‑layer data‑center network topology that replaces traditional three‑tier designs, reducing latency and hops by connecting every leaf switch to every spine switch, thereby improving redundancy, scalability, and cost efficiency for east‑west traffic.
This article introduces the Disruptor in‑memory message queue—its architecture, core concepts such as Ring Buffer, Sequence, Sequencer and Wait Strategies, key features like multicast and gating, tuning options for producers and wait strategies, and provides a complete Java example to help developers build ultra‑low‑latency systems.
Baidu Intelligent Cloud built a 4µs end-to-end low‑latency HPN cluster, optimized traffic management and communication operators, and introduced dynamic expert balancing to dramatically improve the performance of large‑scale PD‑separated inference services, showcasing the deep integration of network infrastructure with AI workloads.
This article introduces the high‑performance in‑memory queue Disruptor, covering its architecture—including ring buffer, sequence, sequencer, and wait strategies—its key features such as multicast events and lock‑free concurrency, and provides practical tuning advice and a complete Java example.
This article provides an in‑depth comparison of Aeron Queue and Chronicle Queue, examining their architectures, ideal use cases, performance characteristics, scalability, and persistence features to help engineers decide which low‑latency messaging system best fits distributed or single‑process applications.
This article introduces the Disruptor in‑memory message queue, explains its core components such as Ring Buffer, Sequence, Sequencer and Wait Strategies, describes its distinctive features like multicast events and lock‑free concurrency, and provides tuning guidelines and a complete Java example.
This article introduces the Disruptor in‑memory message queue, explains its architecture and core concepts such as Ring Buffer, Sequence, Sequencer, and Wait Strategies, and provides tuning guidance and a complete Java example for building low‑latency concurrent systems.
Aeron is an open‑source, low‑latency, high‑throughput messaging framework that leverages zero‑copy memory, shared‑memory IPC and UDP transport to deliver microsecond‑level latency for finance, gaming, and distributed systems, offering a simple API and powerful performance features.
This article explains how to design and implement a high‑performance custom network protocol in Rust, covering zero‑copy parsing, memory‑mapped packet pools, lock‑free event loops, and an efficient binary packet format to achieve ultra‑low latency and massive concurrency.
This tutorial walks through the complete RDMA programming workflow using the ibverbs API, covering device initialization, memory registration, completion queue and queue pair creation, state transitions, send/receive operations, completion handling, and resource cleanup with concrete C code examples.
This article explains Netflix's distributed counter abstraction built on their time‑series service, detailing use cases, API design, counter types, implementation methods, control‑plane configuration, performance results, and future work to achieve near‑real‑time, low‑latency counting at massive scale.
This article explains ZGC’s ultra‑low pause times, massive heap support, colored pointers, concurrent compaction, and generational optimizations, then shows how these features benefit real‑time analytics, high‑performance servers, and online transaction systems, and provides step‑by‑step instructions for enabling and tuning ZGC in Java.
This article provides a comprehensive overview of RocketMQ, covering its architecture, key features such as high reliability, low latency and high throughput, comparisons with Kafka, RabbitMQ and ActiveMQ, and detailed mechanisms that ensure message durability, performance, and ordered consumption.
The article examines the ultra‑large‑scale, high‑bandwidth, low‑latency, and automated network infrastructure needed for training generative AI models, covering custom network designs, congestion control, deterministic RDMA, topology choices such as Fat‑Tree, and emerging deterministic networking technologies.
This article presents JD Advertising's technical practices for improving GPU throughput and reducing latency in large‑scale recommendation scenarios, covering system challenges, storage and compute optimizations for training, low‑latency inference techniques, and compiler extensions to handle massive sparse models.
This article explains the principles, features, and workflow of Java's Z Garbage Collector, provides detailed configuration and logging guidance, and shares AutoMQ's practical tuning experiences that achieve sub‑millisecond pause times and improved performance for latency‑sensitive backend services.
The article analyzes traditional cloud data‑center networking challenges for AI workloads and compares two‑layer and three‑layer fat‑tree architectures, presenting high‑bandwidth, non‑blocking, and low‑latency designs such as AI‑Pool networks and offering practical deployment scales from hundreds to tens of thousands of GPUs.
While Rust’s safety and performance promise significant productivity gains over C++ for low‑latency trading systems, steep learning curves, scarce talent, and lower salaries hinder its adoption in finance, with industry leaders noting both its advantages and the challenges of transitioning legacy codebases.
This article provides a comprehensive technical overview of InfiniBand, covering its rapid adoption in top supercomputers, detailed performance advantages such as ultra‑high bandwidth, CPU offload, sub‑microsecond latency, flexible scalability, QoS, SHARP acceleration, and a comparison with Ethernet, Fibre Channel, and Omni‑Path, while also outlining HDR switch and NIC product families.
mica-mqtt is a lightweight, low‑latency, high‑performance open‑source MQTT component built on Java AIO, offering full MQTT v3.1/v3.1.1/v5 support, WebSocket, REST API, client/server capabilities, clustering via Redis, GraalVM native compilation, Spring Boot starter integration, and extensive customization for IoT and messaging scenarios.
The article details how Quanjian K‑Song has built a comprehensive mobile‑focused audio recording system since 2014, covering low‑latency capture, high‑quality sampling, lyric and vocal‑accompaniment alignment, ear‑return, pitch shifting, vocal enhancement, 3A processing, and AI‑driven scoring to deliver a professional karaoke experience on smartphones.
Quanmin K Song’s decade‑long mobile‑recording platform combines 48 kHz/16‑bit dry‑signal capture, sub‑70 ms latency via OpenSL ES/AAudio, real‑time clipping and noise detection, lyric‑ and vocal‑accompaniment alignment, pitch‑shifting, adaptive vocal enhancement, 3A DSP/AI processing, and AI‑driven pitch correction to deliver industry‑leading high‑quality mobile singing experiences.
An ultra‑large hedge fund with over $10 billion AUM replaced ksqlDB and Flink with RisingWave, leveraging its PostgreSQL‑compatible streaming SQL to achieve sub‑10 ms latency, lower learning and operational costs, rich connectors, advanced operators, and comprehensive observability for real‑time trade data processing.
This article explains how to build a real‑time typeahead (autocomplete) service that returns the most frequent query suggestions using an in‑memory Trie, sharding, offline aggregation, and caching to achieve sub‑20 ms latency, fault tolerance, and horizontal scalability.
An in‑depth analysis of the Tianwang risk‑control Lingji system reveals how JDK 17’s ZGC, Netty’s direct‑memory allocation, and cross‑data‑center channel limits caused severe latency spikes, memory growth, and CPU usage, and outlines the debugging steps and configuration changes that finally resolved the issue.
This article details Tencent Cloud's V265/TXAV1 live streaming solution, covering its high‑efficiency 8K and low‑latency capabilities, performance gains over X265, MV‑HEVC 3D compression, extensive engineering and algorithmic optimizations, and the resulting speed‑up and quality improvements for ultra‑high‑definition live broadcasts.
The article analyzes the limitations of traditional data‑center networking for AI compute workloads and presents high‑bandwidth, non‑blocking, low‑latency design solutions—including two‑layer and three‑layer fat‑tree architectures, AI‑Pool concepts, and recommended configurations—for building scalable, efficient intelligent computing clusters.
At ICASSP 2023, Kuaishou’s audio team presented a low‑latency hierarchical fusion network for full‑band acoustic echo cancellation, detailing its multi‑stage design, asymmetric windowing, loss functions, training strategy, and achieving second place in the non‑personalized AEC Challenge, with real‑world deployment results.
The Dynamic Graph Sampling (DGS) service, built on GraphLearn, delivers sub‑20 ms latency for real‑time GNN inference on large, constantly evolving graphs by separating storage from computation, using event‑driven pre‑sampling, lazy multi‑hop concatenation, and a publish‑subscribe architecture that scales linearly across distributed workers.
ZGC, the Z Garbage Collector introduced in JDK 11, is a low‑latency, region‑based JVM garbage collector that uses colored pointers, load barriers, and multi‑mapping to achieve sub‑10 ms pause times across heap sizes up to several terabytes, with concurrent marking, relocation, and three brief STW phases.
This article details the end‑to‑end low‑latency video streaming architecture, measurement methods, and optimization techniques—including FLV‑2s, RTM, MiniSDP, buffer‑driven speed‑up, and CDN strategies—that Douyin's Volcano Engine video cloud employed to achieve sub‑2‑second latency and high QoE for World Cup live broadcasts.
Bilibili built a cross‑platform cloud‑gaming service that leverages WebRTC with tuned jitter buffers, unordered data channels, adaptive input‑report rates, and a custom kernel driver to deliver sub‑100 ms latency, dynamic bitrate control, and haptic feedback, overcoming typical latency, stutter, and flexibility limitations of existing solutions.
NetEase and China Construction 8th Bureau teamed up to create high‑altitude excavation robots that rely on NetEase Cloud Sign's ultra‑low‑latency global transmission, edge deployment, intelligent scheduling, WE‑CAN routing, multiplexed signaling, and advanced QoS techniques to achieve sub‑150 ms control and reliable video streaming in challenging network environments.
This article examines the challenges of deploying dozens of deep‑learning models at Zuoyebang and compares three serving architectures—Gunicorn + Flask + Transformers, Tornado + PyTorch, and Tornado + Triton—highlighting performance trade‑offs and presenting a final high‑concurrency, low‑latency solution in production.
The article surveys the evolution of live streaming toward low‑latency, interactive scenarios and details WebRTC‑based real‑time audio/video system design, covering RTP/UDP transport, FEC and ARQ loss recovery, congestion control, jitter buffering, echo cancellation, edge‑node path optimization, and a multi‑layer architecture with signaling, routing, mixing services for scalable, high‑availability PK deployments.
This article provides a comprehensive overview of low‑latency architecture, covering network hardware, system‑level programming strategies, language choices, memory management techniques, event‑driven designs, high‑performance data structures, and visualization approaches for building ultra‑fast computing systems.
This article explains how high‑frequency trading firms achieve ultra‑low network latency by combining proximity deployment, dedicated links, microwave transmission, InfiniBand, low‑latency switches, kernel bypass, RDMA, TCP offload engines and FPGA acceleration, and summarizes the impact of each technique on overall request latency.
This article details the investigation of a 2% timeout rate in a billion‑request‑per‑day ticket pricing service, identifies GC‑induced stop‑the‑world pauses as the main cause, and demonstrates how switching from ParNew+CMS to G1 and finally to ZGC dramatically reduces latency and timeout rates.
At the 16th OSDI conference, Tencent’s WeChat team presented the award‑winning Ekko system—a groundbreaking, ultra‑low‑latency model‑update solution for massive recommendation workloads that dramatically speeds up updates, supports over a trillion‑scale models, and has already boosted user engagement across billions of daily users.
In a May 16 live session, Huawei Cloud’s chief architect MuNa explained the goals of media networking, detailed the SparkRTC global network architecture—including overlay/underlay layers and low‑latency techniques—and revealed nine key strategies such as end‑to‑end transmission, AI‑driven routing, multi‑path access, and edge acceleration, while also reminding developers to register for the ongoing competition.
The article analyzes how the AI boom and high‑performance SSD storage demand sub‑microsecond latency, exposing TCP/IP’s inherent context‑switch and CPU overhead, and explains why RDMA’s kernel‑bypass, zero‑copy design and 1 µs latency make it the preferred network stack for modern data‑center workloads despite challenges in Ethernet deployment.
This article explains the fundamentals of Remote Direct Memory Access (RDMA), comparing it with traditional networking, outlining its core advantages, suitable use cases, the three main RDMA protocols (Infiniband, RoCE, iWARP), deployment requirements, communication flow, and essential programming concepts.
This article explains how Aerospike, a high‑performance NoSQL database, achieves millisecond‑level query latency on terabyte‑scale datasets by using a hybrid storage architecture, multi‑level storage tiers, flash optimizations, and a flexible ecosystem that supports diverse real‑time use cases.
The article explains how BIGO RTC achieves high‑quality, low‑cost real‑time audio‑video communication by optimizing core video operators such as HEVC encoding, AI‑driven super‑resolution and HDR, and by employing scene‑adaptive techniques like device performance tuning, content‑adaptive encoding and AI‑based pre‑processing to meet diverse latency constraints.
This article details the design and implementation of a self‑developed interactive live‑streaming platform that supports massive concurrent users and ultra‑low latency for online education, covering business scenarios, technical abstractions, key low‑latency and high‑concurrency techniques, and real‑world performance results.
ZGC (Z Garbage Collector) is a scalable, low‑latency collector introduced in JDK 11, featuring sub‑millisecond pause times, region‑based memory layout, colored pointers, NUMA awareness, and concurrent phases, with detailed explanations of its architecture, parameters, triggers, and log analysis.
ZGC, introduced in JDK 11 and stabilized in JDK 15, is a low‑latency, region‑based, concurrent garbage collector that aims for sub‑millisecond pause times regardless of heap size, leveraging colored pointers, NUMA awareness, multi‑mapping, and read barriers to achieve efficient memory reclamation.
Secure Reliable Transport (SRT) is an open‑source, UDP‑based protocol that overcomes TCP’s latency and jitter issues in long‑distance live streaming by using ARQ retransmission, FEC, and AES/TLS encryption, enabling high‑quality, low‑delay video delivery even in poor network conditions.
The article analyzes how the AI era’s demand for ultra‑low latency and high throughput exposes fundamental limits of the traditional TCP/IP stack, and explains why RDMA’s kernel‑bypass, zero‑copy design, and emerging congestion‑control algorithms are becoming the preferred network fabric for modern data‑center workloads.
This article explores how Tencent Cloud leveraged WebRTC and innovative engineering to create an ultra‑low‑latency live streaming solution that outperforms traditional CDN‑based streams, detailing the technical challenges, product decisions, and market impact that have driven rapid adoption across e‑commerce, education, and entertainment sectors.
The article outlines Vivo's live streaming platform evolution, detailing its business scenarios, underlying architecture, key technical challenges such as beautification, messaging, latency, and cost, and the concrete solutions implemented across entertainment, interactive, and corporate event live streams.
The article analyses the metaverse as a network-driven evolution, explaining its definition, classifying terminal types, and detailing the new security, ultra‑low‑latency, and deterministic quality‑of‑service requirements that challenge existing internet architecture and design principles.
This article details the design and implementation of a low‑latency, high‑definition multi‑angle live streaming solution using WebRTC, covering protocol selection, system architecture, edge commands, client integration, performance optimizations, and lessons learned from deploying the feature in a large‑scale live event.
The mica‑mqtt library, built on t‑io, offers a simple, low‑latency, high‑performance open‑source MQTT solution for IoT, supporting MQTT v3.1/v3.1.1/v5, WebSocket sub‑protocol, HTTP API, clustering via Redis, Spring Boot starter integration, Prometheus/Grafana monitoring, and provides extensive client and server interfaces with code examples.
5G’s high bandwidth, low latency, and edge offloading empower real‑time remote control of complex equipment, demanding synchronized video, audio, and control signals, which are addressed by real‑time AV communication, control‑signal synchronization, and network optimization, while evolving system architectures—from direct‑connection models to integrated fusion designs—ensure scalability, reliability, and minimal latency for applications in mines, ports, logistics, and autonomous vehicles.
This guide introduces mica-mqtt, a simple, low‑latency, high‑performance MQTT open‑source component built on t‑io, outlines its extensive feature set, provides Spring Boot and plain Java integration steps, showcases configuration examples, and explains monitoring and clustering capabilities for IoT back‑end development.
The article examines how FPGA’s pipeline parallel architecture provides latency‑critical advantages over CPU and GPU for both compute‑intensive workloads such as matrix operations and AI inference, and communication‑intensive tasks like encryption and high‑throughput networking, while also discussing deployment models, power efficiency, eFPGA trends, and the evolving Chinese FPGA market.
This article examines Youku's technical team's systematic approach to improving video playback by tackling slow start‑up, error handling, network stalls, audio‑video sync, and overall stability through metric‑driven optimization, adaptive streaming protocols, device capability databases, and forward‑looking features like low latency and high‑frame‑rate streaming.
Intelligent edge computing extends traditional edge architectures by embedding AI-driven data analysis and machine learning directly at the network edge, enabling faster, low‑latency decisions for use cases like smart cameras and autonomous vehicles, and is rapidly becoming a core strategy for modern enterprises.
This article examines the technical challenges of massive e‑commerce live streams, explains why traditional streaming suffers from high latency, and details Huawei Cloud's low‑latency architecture, protocols, intelligent scheduling and ultra‑fast transcoding that enable sub‑second product link delivery to millions of viewers.
Tencent's Kona JDK 11 introduces a production‑ready ZGC implementation that reduces Java garbage‑collection stop‑the‑world pauses to under 10 ms, enabling ultra‑low‑latency online services across massive heaps while maintaining acceptable throughput, and the article details its design, tuning, and real‑world deployments.
This article explains the drawbacks of traditional high‑latency timeout centers and presents a Redis‑based low‑latency design, detailing task storage, scheduling, topic and queue structures, two‑phase consumption, retry control, and the resulting performance and reliability benefits.
WE‑CAN, NetEase Cloud Communication's self‑developed large‑scale distributed transmission network, dramatically improves end‑to‑end data quality, reduces costs, and supports diverse real‑time communication scenarios through intelligent routing, layered decoupling, and global edge deployment, as demonstrated by extensive production metrics.
This article explains the technical challenges of high‑latency video decoding in Android game streaming with Moonlight, analyzes why it occurs, and provides practical solutions—including API choices, immediate buffer release, frame ordering, and low‑latency vendor configurations—to achieve smoother, low‑delay gameplay.
This article explains the fundamentals of RTC, distinguishes it from traditional streaming, highlights WebRTC's cross‑platform, plugin‑free benefits, and shows how low‑latency real‑time communication can enhance interactive applications.
The article explains ZGC, Oracle's low‑latency Java garbage collector introduced in JDK 11, covering its dynamic region layout, colored pointer technique, tri‑color marking, memory multi‑mapping, read barriers, operation phases, performance characteristics, and its main drawback of floating garbage.
Tencent's Real‑Time Audio‑Video (TRTC) platform delivers sub‑second, cross‑platform RTC for up to 300 participants and 100 000 viewers, using UDP‑based NACK/FEC, dynamic QoS, AI‑enhanced video/audio effects, and cloud‑driven QoE optimization to support diverse scenarios such as online meetings, education, live streaming, and industry‑specific applications.
This article explains the fundamentals of RDMA and RoCE, compares RoCE v1 and v2, outlines deployment steps, highlights performance benefits such as low CPU usage and zero‑copy, and answers common questions about its differences from iWARP and InfiniBand, helping data‑center engineers evaluate the technology.
To overcome HBase’s weak availability and GC‑induced latency spikes, the DiDi team introduced a replication‑based client multi‑read (hedged‑read) mechanism and migrated to the Z Garbage Collector, which together dramatically cut maximum and 99.9th‑percentile latencies while keeping services online during region disruptions.
The article explains how a high‑frequency trading firm evaluated Java versus C++ for ultra‑low‑latency trading, discusses the challenges of JVM JIT compilation and garbage‑collection pauses, and shows how Azul Zing’s C4 collector delivers near‑C++ latency while preserving Java’s development productivity.
The ARC Car Cloud Control platform now streams the vehicle’s screen using Android’s Virtual Display and a C++‑based H.264 hardware encoder, sending raw video over a TCP socket to a server that adaptively adjusts bitrate and frame rate, while the web client decodes the fragmented MP4 via MSE, dramatically lowering CPU usage and latency on low‑end head‑units.
This article details Youku's membership marketing system architecture, covering complex marketing scenarios, high‑availability and low‑latency requirements, rule‑based QL engine, unified marketing framework, multi‑cache storage, multithreaded matching, asynchronous reward distribution, and distributed transaction mechanisms.
This article examines the Z Garbage Collector (ZGC) introduced in JDK 11, detailing its low‑pause design goals, underlying concurrent marking‑copy algorithm, colored pointer and read‑barrier techniques, practical tuning parameters, real‑world case studies, and the performance impact of upgrading from CMS/G1 to ZGC in high‑throughput, low‑latency services.
The article explains how Meituan’s risk‑control platform eliminated frequent 40 ms CMS pauses by adopting JDK 11’s ZGC—detailing its concurrent mark‑copy design, practical tuning parameters, real‑world case fixes, and measured latency reductions of up to 74 % while noting trade‑offs.
This article examines the rapid rise of live‑ecommerce during the 618 shopping festival, explains why the “live + ecommerce” model demands robust backend, streaming and CDN infrastructure, and details Tencent Cloud’s architectural solutions—including media processing, low‑latency protocols, bandwidth optimization and anti‑attack measures—to support massive concurrent traffic.
The article explains how the Secure Reliable Transport (SRT) protocol, combined with Tencent Video Cloud’s optimized infrastructure, overcame latency and packet‑loss challenges to deliver stable, high‑quality live streams for the 2020 LPL Mid‑Season Cup and other large‑scale events, and describes its broader applications through MediaConnect.
This article explains the design goals, architecture, key features, tuning options, and version history of Java's Z Garbage Collector (ZGC), highlighting its sub‑10 ms pause times for terabyte‑scale heaps, its use of colored pointers and load barriers, and the trade‑offs in throughput and configuration.
This article analyzes the challenges of low‑latency, stable, and cost‑effective online multiplayer games and explains how Tencent's Game Server Engine (GSE) provides elastic scaling, near‑by scheduling, stateful shrinkage, multi‑region disaster recovery, and zero‑downtime updates to meet those demands.
This article explains how Apache Hudi provides an incremental processing framework that enables efficient, low‑latency data ingestion, storage, and query capabilities on Hadoop, detailing its architecture, storage layout, compaction, write and read paths, and support for real‑time and batch analytics.
In an interview, iQIYI Live Cloud’s Chen Kunzhong explains how their WebRTC‑based RTCDN reduces encoding and transmission delay to achieve roughly 100 ms end‑to‑end latency for cloud‑gaming ads, supporting cross‑device play, instant click‑to‑play sessions, and future 5G‑enhanced high‑resolution experiences.
This article reviews the evolution of Alibaba's Aliware message engine, analyzes the low‑latency and high‑availability challenges faced during Double 11, and details the architectural, JVM, memory, rate‑limiting, and multi‑replica solutions that enabled RocketMQ to achieve sub‑millisecond write latency and five‑nine availability.
This article examines Pharos, Everbright Bank's home‑grown HBase indexing middleware, detailing why existing secondary‑index solutions fall short, the design goals of low latency, simple architecture and non‑intrusiveness, and the concrete storage, pagination, and transaction‑consistency techniques that enable fast complex queries on massive data.
