Replace samples with batches of samples in the data pipeline

[matthias-wende-frequenz]

What's needed?

We have designed the flow of data inside the sdk's data pipeline such that each sample is handled individually. That implies that in each node of a frequenz-sdk data pipeline we're having an async context switch which eventually lead to poor performance.

We can generalize this concept when replacing the role of the Sample in the data pipeline with a batch of samples. When setting the batch size to one we're getting the current approach as a special case.

The advantage is that in certain cases we can fill up a batch with a fixed number of samples.

The batch size might be adaptable depending on e.g. the incoming sample period and the resampling period.

Proposed solution

In a first step we can introduce a batch type that represents a list of samples and update all parts of the data pipeline accordingly.
That opens space for performance optimizations that can be addressed in succeeding steps.

Use cases

• increase the band with of the data pipeline
• reduce number of context switches in e.g. the resampler
• reading historical data from disc