Performance and I/O

Nitrux enables multiple changes in the kernel I/O and memory systems.

• We use asynchronous garbage collection by default to avoid synchronous updates to access or modification times.
• Modify the rate at which the kernel reclaims VFS caches, enabling asynchronous, non-blocking I/O, and reducing the aggressiveness of the kernel’s swapping out anonymous memory relative to pagecache and other caches.
• Increase Linux autotuning TCP buffer limits. These TCP buffer settings enhance high-speed network performance by dynamically adjusting buffer sizes to improve throughput, reduce packet loss, prevent congestion, and ensure smooth data transfers, all while efficiently managing system memory.
• Increase the maximum number of packets queued for processing before dropping. It improves handling of bursty traffic, reduces packet drops, and enhances performance and stability on high-speed networks under heavy load.
• Reuse Time-Wait TCP connections to reduce the number of open TCP connections. Efficiently manage network connections, reducing open TCP connections, freeing up system resources, and improving performance.
• Control the boost applied to the watermark (the minimum number of free pages the kernel tries to maintain). Maintain higher free memory to prevent the system from stalling due to memory shortages.
• Change dirty page settings to reduce I/O disk access. Allow for efficient memory usage without frequent disk access.
• Disable the 32-bit VDSO (Virtual Dynamic Shared Object) on 64-bit systems.
• Enable AMD Preferred Core functionality to ensure the CPU’s most performant cores run scheduled tasks.
• SCX scheduler based on whether the system is running on AC or battery power. The SCX framework uses dispatch queues (DSQs) to manage task queuing and execution, including a global FIFO queue (SCX_DSQ_GLOBAL) and per-CPU local queues (SCX_DSQ_LOCAL).
• Enable the AMD 3D V-Cache Optimizer on supported systems to prioritize the CCD with the larger L3 cache, improving application performance.
• Configure GameMode to use its heuristics to pin and park cores on AMD 3D V-Cache processors and Intel processors with P/E cores, avoiding performance loss from suboptimal core scheduling.
• ADIOS (Adaptive Deadline I/O Scheduler). A block-layer I/O scheduler for the Linux kernel, designed for modern multi-queue block devices (blk-mq), providing low-latency disk access while maintaining fairness across concurrent workloads.
• Expanded file monitoring. Increasing the inotify watch limit prevents “limit reached” errors in applications that monitor vast numbers of files, such as IDEs (VS Code, IntelliJ) and file synchronization tools.
• Enable TCP Fast Open. We enable TCP Fast Open to eliminate the initial handshake delay for supported applications, speeding up repetitive data connections.
• Enable zswap. Compresses memory pages before writing to swap, reducing disk I/O. Zswap is preferred over zram because it retains disk-backed swap as a fallback, providing both performance and stability under heavy memory pressure.
• Optimized RCU settings.
  • Enable expedited RCU grace periods to improve system responsiveness in specific scenarios.
  • Offload all RCU (Read-Copy-Update) callbacks to kernel threads, reducing interrupt contention on CPUs.
  • Enable “lazy” RCU mode to delay some operations and optimize performance.
  • Enable additional security checks for user copy operations.