Efficient formats for TD

[takuki]

For example, WebAssembly uses LLVM which is a platform-independent intermediate code as the efficient format. It is likely the case that EXI or CBOR does not help WebAssembly a lot.

Similarly. when we think about "Efficient format" for TD, I think we need to step back and think about one or two plausible use cases, and any possible approaches to efficient format should cater to the use cases.

There are also consideration of levels.
There are multiple levels from bytes, syntax (token/grammar), data model, information model.

• The byte level is deflate, gzip, etc. They detect redundancy at byte level, and compress them.
• The syntax level is EXI4JSON, CBOR, etc. They work at the token level according to JSON grammar.
• The data model level is RDF Binary Thrift, HDT, etc. They work at the triplet level according to RDF grammar.
• The information model level works on TD information model described in TD specification.

The higher in the levels, it is more useful for the processing at the high layer because the information is exchanged at the high level directly without going through the lower level processing. However, such a format is not useful for software expecting lower level processing. For example, WoT client that processes TD as pure JSON may not be able to process TDs encoded at RDF level.

Therefore, I suggest we start looking at the use case(s) first.

[danielpeintner]

I would like to add a structure based on the devices and/or use-cases pointing to the aforementioned levels.

Why an efficient TD format might make sense:

• powerful devices may want to get large TDs in compressed form (-> byte level)

  • faster download
  • can be achieved by deflate/gzip as done today in Web

• thing with limited resources may prefer compact form (-> syntax level)

  • to hold TD data
  • to receive TDs it interacts with

[takuki]

* thing with limited resources may prefer compact form (-> syntax level)
  
  * to hold TD data
  * to receive TDs it interacts with

SensorML may give us some insights into issues related to Things with limited resources. In particular, it has an encoding examples for EXI and CBOR.

[takuki]

In March 27th TD telecon (minutes here):

• @sebastiankb indicated there is some use cases for efficient TD encoding in eCl@ss discussion.
• @egekorkan wanted to make clear what kind of constrained devices we are talking about here.

[Citrullin]

Not sure how the W3C working groups operates. Just to give you some insights for constrained devices. I contribute to RIOT OS. We mostly use CBOR for efficient and flexible data exchange. Zephyr OS also uses cbor. Same goes for Amazon FreeRTOS. They all don't have an integration for EXI. Never heard about a use-case with EXI or EXI4JSON. Either it is CBOR or JSON.
Due to the nature of lossy networks, such as 6LoWPAN with RPL, CBOR makes more sense for constrained devices.

[sebastiankb]

defer this topic to TD 2.0