0 views collected around this technical thread.
The article explains how network failures such as packet loss, reordering, latency, and ambiguous node failures make distributed systems unreliable, compares synchronous and asynchronous networks, and discusses the trade‑off between timeout settings and resource utilization.
The benchmark comparing MySQL 9.0 and PostgreSQL 17.0 shows PostgreSQL consistently achieving lower insertion latency, roughly double the throughput (≈19 k QPS vs 10 k QPS), using less CPU, memory and disk I/O, while both hit an 80‑connection pool limit and PostgreSQL maintains stable read latency as MySQL’s performance degrades beyond ~5.5 k QPS.
This article describes how Tubi improved the latency of its Redis‑backed online feature store for machine‑learning inference by analyzing query patterns, measuring client‑side bottlenecks, and applying optimizations such as binary Avro encoding, MGET usage, virtual partitioning, and parallel deserialization to meet a sub‑10 ms SLA.
Designing distributed systems often falls prey to misconceptions such as assuming reliable networks, zero latency, unlimited bandwidth, inherent security, static topology, zero transmission cost, and full autonomy, but applying retries, idempotency, message queues, encryption, dynamic discovery, caching, and time protocols can mitigate these issues.
This article investigates why an RPC interface with an average execution time of 1.5 ms still experiences numerous 100 ms+ timeouts, analyzes the root causes such as GC and I/O jitter, and proposes an elastic timeout optimization to improve service reliability.
Redis can appear slow for many reasons—including baseline latency differences, high‑complexity commands, big keys, concentrated expirations, memory limits, fork overhead, huge pages, CPU binding, swap usage, memory fragmentation, and AOF configuration—so this guide explains each cause, how to diagnose it, and practical optimization steps.
The article examines the write‑after‑read consistency problem in multi‑active cross‑region systems, compares single‑write‑single‑read routing, quorum‑based multi‑write‑multi‑read, and read‑while‑copy methods, explains why primary‑secondary replication is preferred, and proposes a four‑step framework—scenario flagging, data marking, latency evaluation, and near‑by asynchronous replication—to meet WAR requirements efficiently.
This article explores critical latency metrics—from nanosecond-level CPU operations to intercontinental network delays—illustrating how each contributes to web performance, and provides real‑world measurements for various connection types, data processing tasks, and user‑perceived response times to help developers optimize user experience.
This article explains why MySQL master‑slave replication can become delayed—including single‑threaded binlog processing, high load, network latency, and configuration issues—and provides detailed troubleshooting steps and practical solutions such as semi‑synchronous replication, hardware upgrades, and configuration tuning.
The article explains that software performance is measured in milliseconds to nanoseconds, describes core hardware components—CPU, caches, and DRAM—shows how cache hierarchy speeds differ, defines a clock cycle as the basic time unit, and provides typical latency figures for operations ranging from a single CPU cycle to a full system reboot.
This article revisits the classic “Fallacies of Distributed Systems” introduced by Peter Deutsch, explaining why assumptions such as reliable networks, zero latency, infinite bandwidth, secure and homogeneous communication are false, and offering practical strategies like retries, caching, batching, and security‑first design to build robust distributed applications.
This article presents a comprehensive overview of large language model inference, describing the prefill and decoding stages, key performance metrics such as throughput, latency and QPS, and detailing a series of system-level optimizations—including pipeline parallelism, dynamic batching, KV‑cache quantization, and hardware considerations—to significantly improve inference efficiency on modern GPUs.
This article presents a comprehensive overview of large language model inference, detailing the prefill and decoding stages, key performance metrics such as throughput, latency and QPS, and a series of system-level optimizations—including pipeline parallelism, dynamic batching, specialized attention kernels, virtual memory allocation, KV‑cache quantization, and mixed‑precision strategies—to improve GPU utilization and overall inference efficiency.
This article explains how to determine whether Redis is truly experiencing latency spikes, outlines a step‑by‑step benchmarking process, identifies common causes such as high‑complexity commands, big keys, memory limits, fork overhead, AOF settings, CPU binding, swap usage, and provides concrete configuration and code examples to resolve each issue.
The article outlines a systematic three‑step workflow—general service diagnostics, Redis‑specific checks, and reproducible load testing—to pinpoint a hot‑key‑driven CPU bottleneck, then evaluates mitigation options such as read‑write separation, pipelining, and an application‑level cache, ultimately showing the cache’s effectiveness in cutting latency and CPU usage.
This article explains the common reasons why Redis performance degrades—such as network latency, high‑complexity commands, big keys, concentrated expirations, memory limits, fork overhead, huge pages, AOF settings, CPU binding, swap usage, and memory fragmentation—and provides detailed optimization and troubleshooting steps to restore low latency.
This article presents a detailed benchmark comparing Redpanda 23.1 and Apache Kafka 3.4.0 (with and without KRaft) across multiple AWS instance types, showing how Redpanda consistently delivers higher throughput and dramatically lower end‑to‑end latency, often outperforming Kafka by 4‑20× even with extra hardware.
This article explains how to benchmark Redis latency, identify common reasons for slowdowns such as high‑complexity commands, big keys, concentrated expirations, memory limits, fork overhead, swap usage, and CPU binding, and provides detailed configuration and operational steps to monitor and resolve each issue.
The article explains common reasons why Redis latency increases, such as intrinsic latency, high‑complexity commands, big keys, expiration spikes, memory limits, fork overhead, huge pages, AOF configuration, CPU binding, swap usage and memory fragmentation, and provides detailed troubleshooting steps and optimization techniques.
The article examines why live‑streaming latency—averaging about five seconds on the studied platform versus three seconds on top e‑commerce services—remains high, pinpointing encoder, transcoding, CDN GOP‑cache, and player buffering as key contributors, and proposes optimizations such as shrinking GOP size, minimizing CDN cache tiers, tightening player buffers, and adopting ultra‑low‑latency protocols like RTS, RTM, or QUIC to achieve sub‑second delays.