MDZ_device_load

Locale	Presentation string
en-GB	Device load monitoring

Overview

Some drivers expose a simple way to assess how much performance is being used. This extension provides a way to monitor this value.

Interactions with other extensions

If MDZ_long_term_stats is supported then a new stream is defined with name load%.
A new sample is generated every time at least one of the supported devices signals a temperature change of at least 1 degree Celsius.
The value type will have the following form:


	{

	    device: [D₀, D₁, ... , D_N-1],

	    load: [L₀, L₁, ... , L_N-1]

	}

Each value L_i is a integer percentage. If a certain device does not support load monitoring, then its L_i value will be null.

Configuration file additions

A new value devLoadInterval is added. Valid values are unsigned integers, representing the number of seconds between successive temperature readings. Default value is 10.
The extension is presented to remote peers only if at least one device supports load monitoring.

New common commands

None.

New monitor commands

probeLoad

Parameters: an array of unsigned integers, each one being the linearized index of the device. Specifying [] is equivalent to specifying all the devices.

Purpose: retrieve device load statistics.

Reply: an array containing as many entries as request. Each entry is an integer percentage. Given a device with linear index i its load percentage will be at the i-th position.
If a device does not support load monitoring, value will be null.

Push: allowed, but no more than one.

New admin commands

None.

Issues

Is this really useful?

Notes: legacy miners have those but drivers (AMD drivers at least) does not report this with sufficient speed to be useful to adjust load to prevent system sluggish.
There's also the problem that this is a very gross-grained metric while there are many things that could constitute load (such as for example used bandwidth).
Solution: ?

Is this incoherent in style with regards to other probes?

Notes: it just pulls out a value for all devices, in contrast with others which allow to probe specific devices. It also maps directly to linear device index.
Solution: yes - the whole point is this is much more specific in scope: a computing device is being used, that computing device will be monitored. Because of this very strict relationship, there's no need to have a more complicated mapping.