IETF 105 SWIF Codec Hackathon

Saturday and Sunday, November 16-17, 2019, starting at 9:00

• Hackathon wiki
• Github swif-codec repo
• Slack discussion forum (ask Vincent)
• Final recap presentation

General topics to be discussed and agreed on

N/A

Hackathon participants

• (CA) Cedric Adjih
• (OA) Oumaima Attia
• (VR) Vincent Roca

Tasks (including completion status)

• Essentially integration of linear system decoding library (from Cedric A.).
• Continue simple_client.c testing application.

SWIF Codec, encoder:

SWIF Codec, decoder:

SWIF Codec unitary test:

• T3.1: - DONE - find a way to add unitary tests.
• T3.2: - ON PROGRESS - add as many unitary tests as possible for the encoder.
• T3.3: add as many unitary tests as possible for the decoder.
• T3.4: add as many unitary tests as possible for the server/client transmission.

C-language server/client application:

• T4.1 - DONE - work on server (i.e., the encoding/sending side): requires to design a trivial packet format (transmission of the FEC OTI can be static on the opposite), set up the UDP connection, design a simple rate control scheme (to avoid saturating the receiver), and use the generic API to create repair symbols.
• T4.2: - ALMOST DONE - work on client (i.e., the decoding/receiving side): set up the UDP connection, use the generic API to manage source/repair symbols, to recover from erased source symbols. Test the whole transmission/reception chain.

This work should enable to provide feedback to the Generic API I-D.

Python wrapper:

• T5.1: - ON PROGRESS - (CA, OA)

Python test and simulation applications:

• T6.1: (CA, OA)

I-D/API updates and open problems

I-D/API fixes done:

• error in decoded_source_symbol_callback) (): does not return anything, so change for void return type.

Active I-D/API fixes that remain to be done:

• TODO: the esi_t is probably inappropriate. In certain situations, it may be a 64-bit long identifer. Could it be a variable length integer (a la QUIC)?
• TODO: do we really need the encoder_reset_coding_window()? This is clearly useful at the decoder (specify a brand new coding window), but we expect most encoders to manage their encoding window continuously, without needing to reset it altogether. A decision needs to be taken: keep it or remove it.
• TODO: during the end of session, at an encoder, should we call the callback each time we free a symbol or not? We need to clarify this.
• Added INVALID_ESI to the API. Needed during session startup (first symbol submission). Use another approach to avoid reserving an ESI value to the INVALID state?
• TODO: swif_encoder_get_coding_window_information() is not appropriate when there's non contiguous symbols (e.g., with re-coding use-cases). We need to decide what to do.

Decisions taken, API and I-D fixes @ IETF105 hackathon

Decisions

I-D/API fixes done:

• build_repair() cannot be used when the buffer is allocated locally. Fixed the error in the API by using a double pointer.

Decisions taken, API and I-D fixes @ IETF104 hackathon

Decisions

Licencing considerations:

• decision to use a BSD licence. Yes
• authors of existing code reused by the SWIF codec (Cedric/Vincent/...) agreed to switch to a BSD licence

SWIF Codec project targets:

• Focus on an RLC codec first (as per RLC I-D), the RLNC codec will be addressed as a second step (it will share a lot of code anyway).
• Design the SWIF Codec + tiny C language server (encoding side)/client (decoding side) application.
• Add a Python wrapper to the SWIF Codec, then Python test and simulation applications.

Strategy

• Work on encoder first, with an operational SWIF encoder + server application. Add unitary tests for the encoder.
• Then work on decoder and test the whole chain: server - client. Add unitary tests for the decoder and the whole chain.
• In parallel, work on Python wrapper and python applications.

I-D/API fixes done:

• clarify that this SWIF-codec is not meant to be thread-safe.
• clarification for buffer init/free in function swif_build_repair_symbol().
• clarification (added examples) for the definition of codepoints
• typo (encoder instead of decoder) in swif_decoder_reset_coding_window (ptr to swif_decoder_t rather than swif_decoder);
• update to the generic control block structure (addition of pointers to the codec specific versions of the API functions).