This guide explains how to unleash Nginx's full potential on multi‑core Linux servers by configuring automatic CPU affinity, increasing worker connections with epoll, raising system file limits, and enabling zero‑copy transmission, resulting in up to ten‑fold throughput gains and dramatically lower CPU usage.
The article recounts the original complex implementation of a four‑layer TCP proxy in Easegress, explains why using separate read/write goroutines and custom buffers caused error‑handling and flow‑control difficulties, and then shows how switching to Go's io.Copy (and its variants) dramatically simplifies the code while preserving performance through zero‑copy techniques.
This article explains the fundamentals of the mmap system call, its internal working mechanism, zero‑copy I/O model, advantages, typical use cases, and detailed usage guidelines—including function prototypes, parameter meanings, mapping steps, and sample code—while also exploring how mmap‑mapped files behave during program crashes.
This article breaks down zero‑copy data transfer techniques—write+read, mmap+write, sendfile, sendfile + SG‑DMA, and splice—showing how they reduce CPU copies and context switches to boost I/O performance in modern operating systems.
This article explains the concept of zero‑copy in Linux, compares the four main system calls—sendfile, mmap, splice and tee—describes their APIs, internal mechanisms, performance characteristics, typical use‑cases and provides practical code examples for high‑performance network programming.
This article explains how mmap and zero‑copy mechanisms, combined with DMA and shared‑memory APIs, can dramatically reduce CPU involvement, context switches, and data copies during file and network I/O, thereby improving system performance for high‑throughput applications.
Zero‑copy eliminates CPU‑mediated data copies between user and kernel spaces by using DMA and memory‑mapping, dramatically improving I/O performance, reducing context switches, and enabling high‑throughput applications such as file servers, Kafka brokers, and Java networking frameworks.
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.
Zero‑copy techniques such as mmap + write, sendfile, and sendfile + DMA scatter/gather reduce CPU‑memory copies and context switches compared with traditional read/write IO, improving performance for high‑throughput systems like RocketMQ and Kafka, while explaining user‑kernel mode, DMA, and their trade‑offs.
Zero‑copy is an optimization technique that eliminates unnecessary memory copies between kernel and user space by leveraging DMA, memory‑mapping and specialized system calls, thereby reducing CPU load, latency and improving throughput for high‑performance networking, storage and multimedia workloads.
This article explains why a Java page took five seconds to load, identifies three bottlenecks involving large BLOB fields and file I/O, and walks through four optimization steps—including lazy loading, buffered streams, memory‑mapped files, and sendFile zero‑copy—that reduce the load time to under one second.
This article explains why a naïve 32 KB‑chunk file transfer incurs excessive context switches and memory copies, and how zero‑copy, PageCache, asynchronous I/O, and direct I/O techniques dramatically reduce overhead and boost throughput for large‑scale data transfers.
This article explains the concept of zero‑copy, its benefits for I/O performance, and how Java NIO, Netty, Kafka, RocketMQ and related APIs such as mmap+write and sendfile implement zero‑copy to avoid unnecessary data copying between kernel and user space.
The article explains how traditional I/O incurs four data copies and context switches, and how DMA, zero‑copy techniques such as sendfile, mmap, splice, and Direct I/O reduce copying and system calls, improving performance for disk‑to‑network data transfers in Linux.
This article explains the concept of zero‑copy I/O, how DMA offloads data movement between memory, disks and network cards, and compares practical implementations such as sendfile, mmap and Direct I/O, highlighting their benefits, limitations and typical use cases in Linux systems.
This article explains how zero‑copy techniques such as DMA, sendfile, mmap and Direct I/O reduce the four data copies and context switches normally required for disk‑to‑network transfers, improving performance for systems like Kafka and MySQL while outlining their trade‑offs.
This article explores Kafka’s performance architecture, covering network and disk bottlenecks, sequential writes, zero‑copy techniques, page cache usage, Reactor‑based networking, batch processing, compression, partition concurrency, and file structures, providing practical optimization methods for high‑throughput streaming applications.
This article explains why zero‑copy techniques such as mmap, sendfile and DMA dramatically reduce CPU copies and context switches in traditional read/write I/O, improving the performance of high‑throughput systems like RocketMQ and Kafka.
