[Zstd] About recommended usage

[hyunsooda]

According to the Zstd description, There are two methods provided: stream and block.
Is there a recommended selection based on the data size to be compressed? (e.g., for the size 100MB, which one is better).
In the first statement in Blocks section,

For compressing small blocks,

Providing a more detailed description or an approximate number would be helpful, especially for newcomers.
Thank you.

[klauspost]

Blocks are only really used as an internal structure. I have never seen them used standalone. And they are limited to 128KB.

This library does not expose any block encoder/decoder.

[hyunsooda]

Hi @klauspost, I meant the stream method to this and block to this (EncodeAll()).

In my local experiment for multiple 100MB data, the two methods do not seem very different. So, which method is preferable based on the origin target size to be compressed?

[hyunsooda]

Yes, in my case the multiple of 100MB data is independent, so it can be compressed independently. So, my original question was, which is better based on the target size to be compressed..? For example, I expected that streaming is preferable when the data size range is within A to B, otherwise EncodeAll is prefereable.

[klauspost]

The output is very close to, if not the exact same.

[hyunsooda]

Got it. Thank you. Do you mean the output contains not only the compression rate but also the compression time?

[klauspost]

Streams will typically encode about twice as fast, and should generally be used for anything above 1MB or so.

[hyunsooda]

Thanks for the valuable information. Let me close.

[hyunsooda]