Performance

Clear Linux* OS is built with optimizations across the whole stack for improved performance. Clear Linux OS achieves its performance through a variety of design decisions and software building techniques.

Overview

The Clear Linux OS philosophy is to do everything with performance in mind. The Clear Linux OS team applies this philosophy in the project’s codebase and operating culture.

Below are some examples of the Clear Linux OS philosophy:

Consider performance holistically.

Performance optimizations are considered across hardware and software. Clear Linux OS shows the performance potential of a holistic approach on Linux, using Intel® architecture with optimizations across the full stack.

Optimize for runtime performance.

In general, Clear Linux OS will trade the one-time cost of longer build time and larger storage footprint for the repeated benefit of improved runtime performance. Clear Linux OS users benefit from the optimized software but aren’t affected by the increased build time because the Clear Linux OS team builds the software before distributing it to Clear Linux OS clients.

Optimize performance for server and cloud use cases first.

Design decisions that optimize performance for server and cloud also benefit other use cases, such as IoT devices and desktop clients.

Clear Linux OS has become well-known for the performance it can deliver. Phoronix publishes Linux performance comparisons that include Clear Linux OS.

Software build toolchain

Clear Linux OS uses many techniques in its software build toolchain to improve software performance, such as aggressive compiler flags and CPU-specific optimizations. If maintained manually, these techniques can become complex to support due to the volume of packages and the potential for technical drift of package performance configurations. The Clear Linux OS team built the autospec tool to manage this complexity and to apply the techniques used in the software build toolchain across the entire project. autospec is available as part of the OS for developers to use when they build their own projects on Clear Linux OS.

Latest versions of compilers and low-level libraries

Clear Linux OS is a rolling release distribution and follows upstream software repositories, including compilers and libraries, for updates. Clear Linux OS includes upstream source-level optimizations as soon as they’re available.

A benchmark approach to compiler performance

Clear Linux OS chooses the compiler used to build each software package on a case-by-case basis to maximize performance. Typically, Clear Linux OS uses the open source GNU Compiler Collection (GCC) with the standard low-level libraries Glibc and libstdc++ for C and C++ programming languages. If there is a performance advantage, Clear Linux OS will build packages with Clang / LLVM.

Clear Linux OS uses patched compilers and low-level libraries for exact control of the software build. Patches include changes that default to more aggressive optimizations or optimizations that haven’t yet been merged upstream.

View the full list of patches in the autospec repositories on GitHub:

Aggressive compiler flags

Clear Linux OS uses aggressive compiler flags to optimize software builds for runtime performance. Some significant flags that Clear Linux OS often implements are:

mtune and march

Options used to tune generated code with optimized instructions for specific CPU types instead of creating generic code for maximum compatibility.

Clear Linux OS defines its minimum hardware requirements to be second-generation Intel® microarchitecture code name Westmere (released in 2010) or later. This enables compiler optimizations that are available only on newer architectures. Whenever possible, Clear Linux OS tunes code for the Haswell generation processors or newer.

Clear Linux OS sets march=westmere and mtune=haswell.

Note

Clear Linux OS doesn’t require Advanced Encryption Standard (AES), so it should run on some Intel CPUs from the first generation of Intel® microarchitecture code name Nehalem (released in 2008). Refer to the recommended minimum system requirements for specific requirements.

O3

The largest preset of compiler options optimizations for performance. O3 favors runtime performance.

View the “Optimize Options” section of the GCC man page for additional information: man gcc

LTO

Link-time optimization that performs an optimization between compiled object files and creation of executable binaries by adding extra information to the compiled object to help the linker.

PGO

Profile guided optimization or field guided optimization performs optimization based on information sampled during the execution of the program.

Compiler flags are set at different levels in the Clear Linux OS build environment:

User flags

The set of default flags used by Clear Linux OS when a user compiles software from source. The flags are exported as system-wide environment variables from the /usr/share/defaults/etc/profile file to the user’s shell by default. These are the standard variables read by the compiler, named *FLAGS, depending on the compiler.

Note

Source code may come with software build systems that override these values. This will cause a difference in expected flags. The Clear Linux OS autospec tooling will attempt to ignore these overrides, but the build system may still need patching. A manual build will not ignore the build system override values if they exist.

Global flags

Compiler flags applied at a global level for all packages. The Clear Linux OS RPM configuration (clr-rpm-config) contains global compiler flags. Search the macros file for global_cflags and search the rpmrc file for optflags. Global compiler flags may be overridden.

Note

Clear Linux OS doesn’t use RPMs to install software. Clear Linux OS distributes software in the form of Bundles. The RPM format is only used during the Clear Linux OS build process as a way to resolve dependencies.

Per-package flags

Compiler flags applied at a per-package level. The package’s autospec repository contains the package-specific compiler flags. Search the .spec file for the section starting with export CFLAGS.

Multiple builds of libraries with CPU-specific optimizations

To fully use the capabilities in different generations of CPU hardware, Clear Linux OS will perform multiple builds of libraries with CPU-specific optimizations. For example, Clear Linux OS builds libraries with Intel® Advanced Vector Extensions 2 (Intel® AVX2) and Intel® Advanced Vector Extensions 512 (Intel® AVX-512). Clear Linux OS can then dynamically link to the library with the newest optimization based on the processor in the running system. Runtime libraries used by ordinary applications benefit from these CPU specific optimizations.

The autospec repository for Python* shows an example of this optimization: https://github.com/clearlinux-pkgs/python3

Kernel

A modern kernel with variants optimized for different platforms

Clear Linux OS is a rolling release distribution that uses the newest upstream Linux kernel. The Linux kernel has frequent updates which can include performance enhancements. It’s a policy of the Clear Linux OS team to try to upstream any performance enhancements in the Linux kernel for all to use.

Clear Linux OS builds different kernel variants for compatibility with specific platforms. For example, kernels meant to run on virtual machines skip support for much of the physical hardware that doesn’t show up in VM environments and will slow down boot.

View the kernel configuration and patches to the default native kernel in the autospec repository: https://github.com/clearlinux-pkgs/linux/

Utility to enforce kernel runtime parameters

The Linux kernel exposes parameters for tuning the behavior of drivers and devices such as certain buffers and resource management strategies. Clear Linux OS uses a small utility, clr-power-tweaks, to set and enforce kernel parameter values weighted towards performance upon boot. View the set performance values by running sudo clr_power --debug.

Operating system

Operating system and software build-time optimizations set the stage for high performance. Decisions made after the installation of Clear Linux OS are equally as important.

CPU performance governor

Clear Linux OS uses the performance CPU governor which calls for the CPU to operate at maximum clock frequency. In other words, P-state P0. The idea behind prioritizing maximum CPU performance is that the faster a program finishes execution, the faster the CPU can return to a low energy idle state. See the CPU Power and Performance documentation for further details.

Restructured boot sequence

To optimize boot speed, Clear Linux OS uses a restructured order for boot processes that minimizes the time services wait on slow operations and the time boot processes wait on each other.

Systemd-bootchart is a tool for graphing the boot sequence and writes logs to a file under /run/log. The tool and corresponding log file make diagnosing slow boot problems easier. All Clear Linux OS systems have systemd-bootchart enabled by default for every boot. systemd-bootchart configuration is non-blocking to not materially slow down boot performance.