These options are intended for device and driver developers in particular. If in doubt do not touch as these values can dramatically affect your machine's performance in the negative direction. In general the default configuration will always be the most performant for your machine setup.
But to answer your questions specifically, here is a break down:
Backoff channel heat break tolerance
- The amount of heat that is tolerated in a specific power channel before the channel is moved to "backoff" mode - i.e. turned off to avoid damage. (a power channel here is a device, although there may be more than one power channel per device). High values here can cause device damage and low values can cause devices to turn off during heavy use.
Consecutive time units to mark a page as cold
- Length of time (in scheduler ticks) before a page in memory is marked as "old" and becomes eligable to be shuffled, paged-out or even deallocated by the kernel from underneath the driver. Lower numbers here can cause severe device performance penalties and higher numbers can cause devices to consume larger amounts of physical RAM than is necessary (slowing down the rest of the system slightly).
Backoff channel heat check interval
- The number of scheduler ticks between checking the Backooff channel heat break tolerance timer. Higher numbers are more precise but take up more CPU time to perform the checks.
Backoff idle utility threshold - the number of scheduler ticks of no activity before a device is moved to the "idle" power mode. Higher numbers cause devices to turn off quicker when not in use.
Backoff forced hot channel resume timeout - the number of scheduler ticks after a device is disabled for being too hot before it is re-enabled.
Time unit in 2^minutes for access pattern - some value N in this field implies that every 2-to-the-power-N minutes the access-pattern device check is performed on the device. This check is a more complicated check to decide how the device is being used in order to select the next power state for the device.
Monitor short history - If set, causes the short history (verbose) log to be written to the event log. This is useful for debugging but causes additional disk access.
Monitor channel power history - If set causes power-channel power history to be recorded to the event log.
Monitor long history - If set causes the long history (statistical) log to be written to the event log.
Free pages threshold to move pages out of pinned channel - The number of free pages that a pinned channel has to have before it becomes "unpinned". This is to do with NUMA.
PFNs query rate per node - this is for NUMA/RAM configurations, are refers to the number of page frame number queries per node that occur for any given NUMA node. This is particularly important for multi-core and larger systems.
Monitor check interval - If set then various timer ticks are recorded to the event log.
Backoff frequency threshold - how often the "backoff" timer is checked.
Backoff low utility threshold - at what point of low usage the backoff timer is automatically fired in order to make use of system idle time.
Memory stats check interval - the number of scheduler ticks between checking memory statistics and writing them to the event log.
Pinned pages number above which to mark a channel as pinned - the number of pages that a channel needs to allocate before the channel becomes marked as a pinned channel (which gives it priority for certain actions).
Maximum hot pages to put channel in low power state - the maximum number of pages which are awaiting a driver response before the device is put into idle mode. This is important for network devices in order to get power gain out of throttling the device when packets are coming into the computer faster than the driver can process them.
Relocation rate per node - this is to do with NUMA access.
Backoff high utility threshold - the threshold of high utility before the backoff timer is queried.
