TuneD documentation: Optimizing system throughput, latency, and power consumption

Configures the ACPI driver.

The only currently supported option is platform_profile, which sets the ACPI platform profile sysfs attribute, a generic power/performance preference API for other drivers. Multiple profiles can be specified, separated by |. The first available profile is selected.

Sets audio cards power saving options. The plug-in sets the auto suspend timeout for audio codecs to the value specified by the timeout option.

Currently, the snd_hda_intel and snd_ac97_codec codecs are supported and the timeout value is in seconds. To disable auto suspend for these codecs, set the timeout value to 0. To enforce the controller reset, set the option reset_controller to true. Note that power management is supported per module. Hence, the kernel module names are used as device names.

Adds options to the kernel command line. This plug-in supports the GRUB 2 boot loader and the Boot Loader Specification (BLS).

Customized non-standard location of the GRUB 2 configuration file can be specified by the grub2_cfg_file option.

The kernel options are added to the current GRUB configuration and its templates. Reboot the system for the kernel option to take effect.

Switching to another profile or manually stopping the tuned service removes the additional options. If you shut down or reboot the system, the kernel options persist in the grub.cfg file and grub environment files.

The kernel options can be specified by the following syntax:

Or with an alternative, but equivalent syntax:

Where suffix can be arbitrary (even empty) alphanumeric string which should be unique across all loaded profiles. It is recommended to use the profile name as the suffix (for example, cmdline_my_profile). If there are multiple cmdline options with the same suffix, during the profile load/merge the value which was assigned previously will be used. This is the same behavior as any other plug-in options. The final kernel command line is constructed by concatenating all the resulting cmdline options.

It is also possible to remove kernel options by the following syntax:

Such kernel options will not be concatenated and thus removed during the final kernel command line construction.

The initrd_add_img=IMAGE adds an initrd overlay file IMAGE. If the IMAGE file name begins with '/', the absolute path is used. Otherwise, the current profile directory is used as the base directory for the IMAGE.

The initrd_add_dir=DIR creates an initrd image from the directory DIR and adds the resulting image as an overlay. If the DIR directory name begins with '/', the absolute path is used. Otherwise, the current profile directory is used as the base directory for the DIR.

The initrd_dst_img=PATHNAME sets the name and location of the resulting initrd image. Typically, it is not necessary to use this option. By default, the location of initrd images is /boot and the name of the image is taken as the basename of IMAGE or DIR. This can be overridden by setting initrd_dst_img.

The initrd_remove_dir=VALUE removes the source directory from which the initrd image was built if VALUE is true. Only 'y', 'yes', 't', 'true' and '1' (case insensitive) are accepted as true values for this option. Other values are interpreted as false.

The skip_grub_config=VALUE does not change grub configuration if VALUE is true. However, cmdline options are still processed, and the result is used to verify the current cmdline. Only 'y', 'yes', 't', 'true' and '1' (case insensitive) are accepted as true values for this option. Other values are interpreted as false.

Sets the CPU governor to the value specified by the governor option and dynamically changes the Power Management Quality of Service (PM QoS) CPU Direct Memory Access (DMA) latency according to the CPU load.

Plug-in for tuning various block device options. This plug-in can also dynamically change the advanced power management and spindown timeout setting for a drive according to the current drive utilization. The dynamic tuning is controlled by the dynamic and the global dynamic_tuning option in tuned-main.conf.

The disk plug-in operates on all supported block devices unless a comma separated list of devices is passed to it.

The elevator option sets the Linux I/O scheduler.

The scheduler_quantum option only applies to the CFQ I/O scheduler. It defines the number of I/O requests that CFQ sends to one device at one time, essentially limiting queue depth. The default value is 8 requests. The device being used may support greater queue depth, but increasing the value of quantum will also increase latency, especially for large sequential write work loads.

The apm option sets the Advanced Power Management feature on drives that support it. It corresponds to using the -B option of the hdparm utility. The spindown option puts the drive into idle (low-power) mode, and also sets the standby (spindown) timeout for the drive. It corresponds to using -S option of the hdparm utility.

The readahead option controls how much extra data the operating system reads from disk when performing sequential I/O operations. Increasing the readahead value might improve performance in application environments where sequential reading of large files takes place. The default unit for readahead is KiB. This can be adjusted to sectors by specifying the suffix 's'. If the suffix is specified, there must be at least one space between the number and suffix (for example, readahead=8192 s).

The disk readahead value can be multiplied by the constant specified by the readahead_multiply option.

Dynamically sets the front-side bus (FSB) speed according to the CPU load. This feature can be found on some netbooks and is also known as the Asus Super Hybrid Engine. If the CPU load is lower or equal to the value specified by the load_threshold_powersave option, the plug-in sets the FSB speed to the value specified by the she_powersave option. If the CPU load is higher or equal to the value specified by the load_threshold_normal option, it sets the FSB speed to the value specified by the she_normal option. Static tuning is not supported and the plug-in is transparently disabled if the hardware support for this feature is not detected.

Allows tuning of IRQ affinities, and thus re-implements functionality already present in the scheduler plugin. However, this plugin offers more flexibility, as it allows tuning of individual interrupts with different affinities. When using the irq plugin, make sure to disable IRQ processing in the scheduler plugin by setting its option irq_process=false.

The plugin handles individual IRQs as devices and multiple plugin instances can be defined, each addressing different devices/irqs. The device names used by the plugin are irq<n>, where <n> is the IRQ number. The special device DEFAULT controls values written to /proc/irq/default_smp_affinity, which applies to all non-active IRQs.

The option affinity controls the IRQ affinity to be set. It is a string in "cpulist" format (such as 1,3-4). If the configured affinity is empty, then the affinity of the respective IRQs is not touched.

The option mode is a string which can either be set (default) or intersect. In set mode the affinity is always written as configured, whereas in intersect mode, the new affinity will be calculated as the intersection of the current and the configured affinity. If that intersection is empty, the configured affinity will be used.

Plug-in for irqbalance settings management. The plug-in configures CPUs which should be skipped when rebalancing IRQs in /etc/sysconfig/irqbalance. It then restarts irqbalance if and only if it was previously running.

The banned/skipped CPUs are specified as a CPU list via the banned_cpus option.

Example: The scheduler definition

group.ksoftirqd=0:f:2:*:^\[ksoftirqd

is translated to the kthread definition

group.ksoftirqd=0:f2:*:ksoftirqd.*

Plug-in for applying custom kernel modules options.

This plug-in can set parameters to kernel modules. It creates /etc/modprobe.d/tuned.conf file. The syntax is module=option1=value1 option2=value2…​ where module is the module name and optionx=valuex are module options which may or may not be present.

Modules can also be forced to load/reload by using an additional +r option prefix.

The +r switch will also cause TuneD to try and remove netrom module (if loaded) and try and (re)insert it with the specified parameters. The +r can be followed by an optional comma (+r,) for better readability.

When using +r the module will be loaded immediately by the TuneD daemon itself rather than waiting for the OS to load it with the specified parameters.

Enables or disables barriers for mounts according to the value of the boolean option disable_barriers. The option additionally allows the special value force, which disables barriers even on mountpoints with write back caches.

Configures network driver, hardware and Netfilter settings. Dynamic change of the interface speed according to the interface utilization is also supported. The dynamic tuning is controlled by the dynamic and the global dynamic_tuning option in tuned-main.conf.

A plug-in for avoiding inter-processor interrupts caused by enabling or disabling static keys.

The plug-in has no options; when included, TuneD will keep an open socket with timestamping enabled, thus keeping the static key enabled.

Allows tuning of scheduling priorities, process/thread/IRQ affinities, and CPU isolation.

To prevent processes/threads/IRQs from using certain CPUs, use the isolated_cores option. It changes process/thread affinities, IRQs affinities and it sets default_smp_affinity for IRQs. The CPU affinity mask is adjusted for all processes and threads matching ps_whitelist option subject to success of the sched_setaffinity() system call. The default setting of the ps_whitelist regular expression is .* to match all processes and thread names. To exclude certain processes and threads use ps_blacklist option. The value of this option is also interpreted as a regular expression and process/thread names (ps -eo cmd) are matched against that expression. Profile rollback allows all matching processes and threads to run on all CPUs and restores the IRQ settings prior to the profile application.

Multiple regular expressions for ps_whitelist and ps_blacklist options are allowed and separated by ;. Quoted semicolon \; is taken literally.

The irq_process option controls whether the scheduler plugin applies the isolated_cores parameter to IRQ affinities. The default value is true, which means that the scheduler plugin will move all possible IRQs away from the isolated cores. When irq_process is set to false, the plugin will not change any IRQ affinities.

The default_irq_smp_affinity option controls the values TuneD writes to /proc/irq/default_smp_affinity. The file specifies default affinity mask that applies to all non-active IRQs. Once an IRQ is allocated/activated its affinity bitmask will be set to the default mask.

The following values are supported:

To adjust scheduling policy, priority and affinity for a group of processes/threads, use the following syntax.

Here, rule_prio defines internal TuneD priority of the rule. Rules are sorted based on priority. This is needed for inheritence to be able to reorder previously defined rules. Equal rule_prio rules should be processed in the order they were defined. However, this is Python interpreter dependant. To disable an inherited rule for groupname use:

sched must be one of: f for FIFO, b for batch, r for round robin, o for other, * do not change.

affinity is CPU affinity in hexadecimal. Use * for no change.

prio scheduling priority (see chrt -m).

regex is Python regular expression. It is matched against the output of:

Any given process name may match more than one group. In such a case, the priority and scheduling policy are taken from the last matching regex.

The scheduler plug-in uses perf event loop to catch newly created processes. By default it listens to perf.RECORD_COMM and perf.RECORD_EXIT events. By setting perf_process_fork option to true, perf.RECORD_FORK events will be also listened to. In other words, child processes created by the fork() system call will be processed. Since child processes inherit CPU affinity from their parents, the scheduler plug-in usually does not need to explicitly process these events. As processing perf events can pose a significant CPU overhead, the perf_process_fork option parameter is set to false by default. Due to this, child processes are not processed by the scheduler plug-in.

The CPU overhead of the scheduler plugin can be mitigated by using the scheduler runtime option and setting it to 0. This will completely disable the dynamic scheduler functionality and the perf events will not be monitored and acted upon. The disadvantage ot this approach is the procees/thread tuning will be done only at profile application.

The scheduler plug-in supports process/thread confinement using cgroups v1.

cgroup_mount_point option specifies the path to mount the cgroup filesystem or where TuneD expects it to be mounted. If unset, /sys/fs/cgroup/cpuset is expected.

If cgroup_groups_init option is set to 1 TuneD will create (and remove) all cgroups defined with the cgroup* options. This is the default behavior. If it is set to 0 the cgroups need to be preset by other means.

If cgroup_mount_point_init option is set to 1, TuneD will create (and remove) the cgroup mountpoint. It implies cgroup_groups_init = 1. If set to 0 the cgroups mount point needs to be preset by other means. This is the default behavior.

The cgroup_for_isolated_cores option is the cgroup name used for the isolated_cores option functionality. For example, if a system has 4 CPUs, isolated_cores=1 means that all processes/threads will be moved to CPUs 0,2-3. The scheduler plug-in will isolate the specified core by writing the calculated CPU affinity to the cpuset.cpus control file of the specified cgroup and move all the matching processes/threads to this group. If this option is unset, classic cpuset affinity using sched_setaffinity() will be used.

cgroup.cgroup_name option defines affinities for arbitrary cgroups. Even hierarchic cgroups can be used, but the hieararchy needs to be specified in the correct order. Also TuneD does not do any sanity checks here, with the exception that it forces the cgroup to be under cgroup_mount_point.

The syntax of the scheduler option starting with group. has been augmented to use cgroup.cgroup_name instead of the hexadecimal affinity. The matching processes will be moved to the cgroup cgroup_name. It is also possible to use cgroups which have not been defined by the cgroup. option as described above, i.e. cgroups not managed by TuneD.

All cgroup names are sanitized by replacing all all dots (.) with slashes (/). This is to prevent the plug-in from writing outside cgroup_mount_point.

Option cgroup_ps_blacklist allows excluding processes which belong to the blacklisted cgroups. The regular expression specified by this option is matched against cgroup hierarchies from /proc/PID/cgroups. Cgroups v1 hierarchies from /proc/PID/cgroups are separated by commas ',' prior to regular expression matching. The following is an example of content against which the regular expression is matched against: 10:hugetlb:/,9:perf_event:/,8:blkio:/

Multiple regular expressions can be separated by semicolon ';'. The semicolon represents a logical 'or' operator.

Kernels 5.13 and newer moved some sched_ and numa_balancing_ kernel run-time parameters from /proc/sys/kernel, managed by the sysctl utility, to debugfs, typically mounted under /sys/kernel/debug. TuneD provides an abstraction mechanism for the following parameters via the scheduler plug-in: sched_min_granularity_ns, sched_latency_ns, sched_wakeup_granularity_ns, sched_tunable_scaling, sched_migration_cost_ns, sched_nr_migrate, numa_balancing_scan_delay_ms, numa_balancing_scan_period_min_ms, numa_balancing_scan_period_max_ms and numa_balancing_scan_size_mb. Moreover in kernel 6.6 and newer support for the sched_wakeup_granularity_ns and sched_latency_ns were removed. The sched_min_granularity_ns was renamed to sched_base_slice_ns. Based on the kernel used, TuneD will write the specified value to the correct location or ignore it. For the compatibility the alias sched_base_slice_ns was added, but the sched_min_granularity_ns can be still used instead.

Executes an external script or binary when the profile is loaded or unloaded. You can choose an arbitrary executable.

TuneD calls the executable with one of the following arguments:

You need to correctly implement the stop action in your executable and revert all settings that you changed during the start action. Otherwise, the roll-back step after changing your TuneD profile will not work.

Bash scripts can import the /usr/lib/tuned/functions Bash library and use the functions defined there. Use these functions only for functionality that is not natively provided by TuneD. If a function name starts with an underscore, such as _wifi_set_power_level, consider the function private and do not use it in your scripts, because it might change in the future.

Specify the path to the executable using the script parameter in the plug-in configuration.

Tunes options for SCSI hosts.

The plug-in sets Aggressive Link Power Management (ALPM) to the value specified by the alpm option. The option takes one of three values: min_power, medium_power and max_performance.

Plug-in for tuning SELinux options.

SELinux decisions, such as allowing or denying access, are cached. This cache is known as the Access Vector Cache (AVC). When using these cached decisions, SELinux policy rules need to be checked less, which increases performance. The avc_cache_threshold option allows adjusting the maximum number of AVC entries.

Plug-in for handling sysvinit, sysv-rc, openrc and systemd services.

The syntax is as follows:

Supported service-handling commands are start, stop, enable and disable. The optional file:file directive installs an overlay configuration file file. Multiple commands must be comma (,) or semicolon (;) separated. If the directives conflict, the last one is used.

The service plugin supports configuration overlays only for systemd. In other init systems, this directive is ignored. The configuration overlay files are copied to /etc/systemd/system/service_name.service.d/ directories. Upon profile unloading, the directory is removed if it is empty.

With systemd, the start command is implemented by restart in order to allow loading of the service configuration file overlay.

Sets various kernel parameters at runtime.

This plug-in is used for applying custom sysctl settings and should only be used to change system settings that are not covered by other TuneD plug-ins. If the settings are covered by other TuneD plug-ins, use those plug-ins instead.

The syntax for this plug-in is key=value, where key is the same as the key name provided by the sysctl utility.

Sets various sysfs settings specified by the plug-in options.

The syntax is name=value, where name is the sysfs path to use and value is the value to write. The sysfs path supports the shell-style wildcard characters (see man 7 glob for additional detail).

Use this plugin in case you need to change some settings that are not covered by other plug-ins. Prefer specific plug-ins if they cover the required settings.

Plug-in for tuning systemd options.

The cpu_affinity option allows setting CPUAffinity in /etc/systemd/system.conf. This configures the CPU affinity for the service manager as well as the default CPU affinity for all forked off processes. The option takes a comma-separated list of CPUs with optional CPU ranges specified by the minus sign (-).

An Intel-specific plug-in for limiting the maximum and minimum uncore frequency.

The options max_freq_khz, min_freq_khz correspond to sysfs files exposed by Intel uncore frequency driver. Their values can be specified in kHz or as a percentage of their configurable range.

Sets autosuspend timeout of USB devices to the value specified by the autosuspend option in seconds. If the devices option is specified, the autosuspend option applies to only the USB devices specified, otherwise it applies to all USB devices.

The value 0 means that autosuspend is disabled.

Sets various power saving features on video cards. Radeon cards are supported. The powersave level can be specified by using the radeon_powersave option. Supported values are:

For additional detail, see KMS Power Management Options.

Mobile hardware with amdgpu driven eDP panels can be configured with the panel_power_savings option. This accepts a value range from 0 to 4, where 4 is the highest power savings but will trade off color accuracy.

Tunes selected sysctl options in /proc/sys/vm, currently dirty_ratio, dirty_background_ratio, dirty_bytes, and dirty_background_bytes. See https://docs.kernel.org/admin-guide/sysctl/vm.html for detailed documentation of these options.

Additionaly enables or disables transparent huge pages depending on the value of the transparent_hugepages option. The option can have one of three possible values: always, madvise and never.

The transparent_hugepage.defrag option specifies the defragmentation policy. Possible values for this option are always, defer, defer+madvise, madvise and never. For a detailed explanation of these values refer to Transparent Hugepage Support.