Gleam Lab · Blog Archive
Blog Page 3
Technical exploration and engineering notes, 655 articles in total.
Guava Cache: From ConcurrentHashMap to LocalCache Structure
Deep dive into LocalCache, Guava Cache's core implementation class, covering segmented lock design, five queue mechanism, eviction strategy and lazy cleanup principles.
Cache Practice: Local vs Distributed (Guava/Redis 7.2)
High concurrency read-heavy business, database can't handle it, need to improve throughput and stability.
AI Research #135: Gemini 3 Pro Back on Top - MoE, Million-token Context and Deep Think
Explains Gemini 3 Pro's advantages through sparse MoE architecture, million-token context, native multimodal (text/image/video/PDF), thinking depth control (thinking_leve...
OSS Access Control: ACL, RAM, Bucket Policy and Troubleshooting
For multi-team and cross-account collaboration, OSS needs public access, least privilege, and auditability.
AI Research #134: Java 2025 - Will It Decline?
In 2025, Java remains the enterprise backend and critical industry workhorse.
OSS Practice Guide: Bucket/External Links/Hotlink Protection
Enterprises and individuals migrating images/audio/video/static resources to Aliyun OSS need stable operations and cost control.
AI Research #133: Java vs Kotlin/Go/Rust/Python/Node 2025
Comprehensive comparison of Java 21/23 (Virtual Threads/Loom) with Kotlin 2.x, Go 1.22/1.23, Rust 1.7x, Python (PyTorch 2.x/TensorFlow 2.x), Node.js 20/22.
Java Integration with FastDFS: Client and Maven/Spring Boot
Two ways for Java applications to integrate with FastDFS: official source compilation (happyfish100/fastdfs-client-java 1.37-SNAPSHOT) and Maven dependency (cn.
AI Research #132: Java Ecosystem 2025 - Spring, Quarkus, Micronaut and AOT Trends
Spring Framework 6, Spring Boot 3.x, GraalVM Native Image, Quarkus, Micronaut, and AOT trends shaping the Java ecosystem in 2025.
FastDFS High Concurrency Optimization: max_connections, Trunk Files and nofile
Scenario: High concurrency / massive small files (trunk) / single disk or RAID Conclusion: Upgrade to v5.04+, increase max_connections and align with nofile
AI Research #131: Java 17/21/25 Complete Comparison
Java 17 (2021), Java 21 (2023), Java 25 (2025) language and JVM changes, covering Virtual Threads (Project Loom), Records/Pattern Matching (Project Amber).
FastDFS Cluster Practice: 3-Node Tracker/Storage + Nginx
Deploy tracker, storage, and Nginx on three hosts with round-robin upload. Key points: Nginx+mod_fastdfs routing and monitoring verification.
AI Research #130: Qwen2.5-Omni Practical Applications
Office assistant, education and training, programming and operations, search-enhanced RAG, device control/plugin agents, and companion entertainment.
AI Research #129: Qwen2.5-Omni-7B Key Specs - VRAM, Context and Deployment
Runs stably at FP16 ~14GB VRAM, with INT8/INT4 quantization (<4GB) enabling deployment on consumer GPUs or edge devices.
AI Research #128: Qwen2.5-Omni Training Pipeline - Three-stage Multi-modal Training
Complete training pipeline breakdown for Qwen2.5-Omni: Thinker based on Qwen2.5, vision initialized from Qwen2.5-VL, audio from Whisper-large-v3.
AI Research #127: Qwen2.5-Omni Deep Dive - Thinker-Talker Dual-core Architecture
Engineering breakdown of Qwen2.5-Omni (2024-2025) Thinker-Talker dual-core architecture: unified Transformer decoder for text/image/video/audio fusion, TMRoPE.
AI Research #126: Tesla FSD V14 Engineering - High-Precision Occupancy and World Modeling
Tesla developed AI-driven high-precision occupancy network, converting multi-camera video into continuous 3D Signed Distance Field (SDF).
AI Research #125: Tesla FSD Business Model and Competitive Landscape
As of end 2022, Tesla had ~$2.9 billion in FSD-related deferred revenue Q4 2022 recognized $324 million in FSD revenue
AI Research #124: Tesla FSD V14 Deep Analysis
Tesla FSD V14 real-world performance and road tests, comparing V13.2 on urban roads and highways: key disengagement metrics, lane changes/ramps, destination arrival...
AI Research #123: FSD V14 Deep Analysis - Vision-Only SDF vs V12
3D environment reconstruction Precision: 10cm (3× improvement over V12's ~33cm resolution) Multi-frame spatiotemporal fusion for dynamic object tracking