find out the computational load of the algorithm by measuring some quantity IN TK1 board [quantity?...

Robotics & Edge Computing Jetson & Embedded Systems Jetson TK1
1

In the TK1 board, I am running an algorithm whose computational load (i.e. amount of work the algorithm does) depends on the data it receives at an input. For e.g., suppose the algorithm is receiving light intensity from a sensor. If the light intensity does not change, the algorithm will not compute anything. That is, its computational load is minimum.
If the light intensity changes, the algorithm will compute a n-point FFT of the light signal where the value of n depends on the change (more change will require larger n). That is, the computational load of the algorithm varies with the data.

We want to find out the computational load of the algorithm by measuring some quantity (e.g. voltage, current) in the TK1 board. What is this quantity? We want this quantity to be:
Sensitive (i.e. the quantity should change as soon as the computational load changes), and
Specific (i.e. the quantity should change enough for the change to be clearly detectable when measured and it should not change due to reasons other than computational load).

Thanks in advance!

2

which quantity can measure in TK1 which shows there might be change in an algorithm according to computational load.

3

You may try tegrastats:

sudo /home/ubuntu-or-nvidia/tegrastats

By default the probing period should be 1s, but you may adjust with --interval in ms.

4

Thanks for the reply.
Is there any mathematical formula available which makes mathematical relation which relate some quantity proportional to a load of an algorithm.

5

I don’t think there is a generic answer to this.
It mostly depends on the algorithms that are used and is thus application specific.
What you can observe with tegrastats (or directly with sysfs) is the resource usage.
What resource usage means for the algorithm depends on the algorithm itself, but you may be able to identify bootleneck or some mode changes, if this really changes resources usage.
In some cases the memory access/usage can change, if GPU computation is involved you may see GR3D(GPU) increasing or decreasing, or CPUs…

6

Thanks for the reply.

I think according to your statement, CPU,GPU,RAM quantity will effect according to load of an algorithm.

7

You may check RAM amount usage (and swap if any) as well as memory controler (EMC) used bandwidth, and if you are using video HW encoder/decoder also check NVENC or NVDEC (name may change depending on Jetson board and L4T version).