Fei has over 10 years of development experience at Tencent and Sogou. Through this account, he shares his deep insights on performance.
This article explains why Linux gives each thread its own stack, describes the historical evolution from LinuxThreads to NPTL, and walks through the glibc implementation that allocates, sizes, and releases thread‑stack memory using mmap and related kernel calls.
This article explains the hardware basis of NUMA, how Linux reads ACPI SRAT and SLIT tables to discover CPU‑memory topology, the kernel functions that initialize NUMA structures, and how the memblock allocator incorporates this information to enable performance‑optimizing tools like numactl.
Linux does not expose all physical memory to users; this article explains why the kernel reserves memory, detailing the early memblock allocator, crash kernel reservations, page structure overhead, and the handoff to the buddy system, illustrating how these mechanisms consume several hundred megabytes of RAM.
This article explains how software developers can use the Linux dmidecode command to retrieve detailed server memory hardware information, including size, type, speed, bit width, ECC, channel placement, and rank, and interprets the output to better understand the underlying hardware.
This article explains how the Linux kernel detects physical memory during boot by leveraging the E820 mechanism, where firmware reports memory ranges via interrupt 15H, enabling the kernel to map usable memory addresses for subsequent allocation.
This article explains the Linux kernel's four‑level page‑table mechanism, the performance impact of TLB misses, why using 2 MB HugePages improves address‑translation efficiency, and provides step‑by‑step instructions for reserving and allocating HugePages via boot parameters, sysfs, and mmap.
The article explains the meaning of a process being in the “D” state on Linux, describing it as an uninterruptible sleep caused by blocked I/O, how to identify it with tools like ps or top, why it cannot be killed, and common troubleshooting steps.
This article explains the inner workings of the classic strace command by walking through a handcrafted implementation that uses ptrace to attach to a target process, capture its system calls, read the ORIG_RAX register, and print the syscall name, while also detailing the relevant kernel source code.
This article explains how the SIGPIPE signal is generated in Linux, why it can cause Go‑Rust services to crash without a core dump, and how to prevent it by configuring the signal handler to ignore SIGPIPE.
netcap is an open‑source eBPF‑driven kernel network packet capture tool that extends tcpdump syntax to trace skb‑related functions across the Linux network stack, offering detailed packet tracing, customizable filters, multi‑trace aggregation, and user‑defined output to improve debugging of packet loss and performance issues.
