As some of you might have noticed, the proposal has been delayed a bit. Taking into account the received feedback, we’ve decided to re-focus the SPE to better align with the needs of the ecosystem.
This new approach takes lessons from Studio but doesn’t try to replicate it; Studio already has an awesome video offering with a mature SDK to go along with it. FrameWorks will focus on Livepeer’s core advantage: cost-effectiveness. We can supplement the ecosystem with an E2E service that is very lean and help entities replicate this setup, minimizing operating costs even further.
We’ve updated the proposal to reflect this change in scope. Let us know your thoughts!
Re: the technical roadmap & initial priorities
We totally agree that proper testing and an experimental fork make sense for some of these features before merging them into mainline. Running our own Gateway nodes will also give us better insights into Orchestrator selection and swapping, which we’re keen to improve.
Doug also raised some good points about ensuring that our roadmap supports network demand and operator usability. The initial development roadmap still remains the same, just to give a brief explanation as to why these initial items were chosen:
-
Gateway Documentation: Many operators don’t realize how flexible the HTTP endpoint is for transcode jobs. Also, we want to provide a short crash course on the media landscape, helping new Gateway operators understand core concepts like codecs, containers, frame types, and transport protocols.
-
AV1 codec support: Even though this might not see a lot of demand out of the gate and the network is currently not that optimized for VoD workflows, AV1 is just awesome for that use case and nice to start supporting for future proofing. Among the major video codecs (H264, H265, VP9, and AV1), it’s the only one currently missing.
Starting with devices from 2023+ AV1 hardware decoding is standard and as late as 2024 did browsers actually start adding support for it. These were both major barriers for adoption, so we fully expect this codec to grow in demand. Since Brad already completed most of the work, this is a low-hanging fruit to round out network capabilities. -
Netint support: Probably the item with the least impact, as not a lot of people run Netint cards… However, adding support is low effort now that LPMS runs on a more modern FFmpeg base. The feature was already completed & merged in 2022, except for updates to the build pipelines. We’ll be ordering one of the newer gens Netint cards to do some testing and see if it’s compatible. The first gens were a real PITA to work with, but these cards are very cost-effective to provide transcode capacity at scale.
-
QSV: Enables ‘proper’ CPU transcoding! Very nice for transcoding pools or any Orchestrator looking for cheap remote machines. Also great at doing AV1 transcodes, QSV has been really impressive in capacity & quality on newer devices (2022+?).
-
Smarter session limits: a common pain point for O/T’s. The static session limit and benchmarking tool aren’t well suited for real workloads, sometimes overloading the decoder/encoder chips.
Note that, with the added scope for the E2E media pipeline, development on some of these items might roll over into the followup proposal.