How to combine RX/TX jitter, latency and packet loss in single value?

[Yuriy P.]

Hi Community,

It is very general questions, but I hope that networking experts from the community will be able to help.

I'm looking for proper/common way how to combine RX/TX jitter, latency and packet loss (each metric separately) in single values.

Among the options for jitter are Root Square Sum or simple MAX function, but it would not probably apply to packet loss as it is %.

Examples:

• Jitter = √(Jitter RX² + Jitter TX²)
• Jitter = MAX(Jitter RX, Jitter TX)
• Latency = MAX(Latency RX, Latency TX)
• Latency = Latency RX + Latency TX

Please advice.

[Joseph W. Doherty]

Unsure there is any "proper/common" way to do so.

The closest thing that comes to mind is something like MOS, and in that reference you'll find "The MOS is subject to certain mathematical properties and biases. In general, there is an ongoing debate on the usefulness of the MOS to quantify Quality of Experience in a single scalar value."

[Yuriy P.]

Thanks Joseph and sorry for confusion. I've updated initial post.

[Joseph W. Doherty · Accepted solution]

Ah, think I understand now.  I was wondering about "combine RX/TX", but I didn't ask, as I first thought your were trying to create some overall combined quality metric (for jitter, latency and packet loss) for two way communication.

I'm unsure I see the value of combining RX and TX, as "direction" is very important for all these metrics, and however you choose to combine them is likely to "dilute" their value.

For example, if I have 50 ms latency for both RX and TX, the symmetrically is likely good, but if RX latency is 10 ms and TX latency is 90 ms, I would wonder what's the cause, but if you somehow combine the values, will such a difference still be apparent?

Basically, combining values, takes you down the path seen in basic statics, such as using mean, medium, mode, variance and standard deviation because one value, alone, doesn't often well represent the data being analyzed.  Unlike statics whose datasets often run into many more than two values, TX/RX, again unclear the benefit you're hoping to find.  Unsure anyone will be able to offer a "proper/common way".  As the other posters have already described, certainly you might be able to find some calculation valuable to your "value" of these attributes, but also again, unsure there's a "proper/common" way.

[Yuriy P.]

Thanks again. Your train of thought has convinced me that using the MAX (or even SUM) function is probably the best way to go if we need to highlight a potential network communication problem.

[M02@rt37]

Hello @Yuriy P.,

We could assign weights to each metric based on their relative importance. These weights are subjective and depend on your specific needs and priorities. Then, multiply each metric by its respective weight and sum them up to obtain a combined value. For packet loss, multiply the converted decimal value by its weight.

Assume we have these metrics:

Jitter (RMS): 5 ms, Latency: 20 ms, Packet Loss: 3% (converted to 0.03)

And let's assign the following weights:

Jitter weight: 0.4, Latency weight: 0.4, Packet Loss weight: 0.2.

The combined value can be calculated:

Combined Value = (Jitter * Jitter weight) + (Latency * Latency weight) + (Packet Loss * Packet Loss weight) = (5 * 0.4) + (20 * 0.4) + (0.03 * 0.2) = 2 + 8 + 0.006 = 10.006

In this example, the combined value is 10.006, representing the overall network performance based on the weighted combination of the individual metrics.

[M02@rt37]

@Yuriy P. 

Assign weights to each metric based on their relative importance to your specific application or system. For example, if low latency is critical for your application, you might assign a higher weight to latency compared to packet loss or jitter.

[Yuriy P.]

Thanks for answer! 

It would work if I need to combine every metric type in single value, but I need to combine RX/TX metrics for jitter, latency and packet loss separately.

Sorry for confusion, I've updated the initial post.