Moving towards competitive market pricing

The Livepeer network has been successfully transcoding live and on demand video to the scale of 2M+ minutes processed/week for years. Orchestrators compete for this work on a number of vectors, including availability, latency, and consistency of performance. To date, the price that the orchestrators have charged for work has not been a large input into their ability to attract work, as the largest broadcaster on the network - the Livepeer Studio hosted gateway - has advertised the price it would pay in ETH, and any O who set their price lower, would be eligible for work.

As the usage of the network, and the diversity of apps have grown - from web3 social live streaming, to creator economy, to VOD uploads, to web2 scaled applications - users require different pricing for different services. VOD is cheaper than Live, which requires realtime performance, and different regions can actually come with different costs for node operators. The vision of the Livepeer network is that it is the world’s most cost effective infrastructure network for video, because of the open market competition between node operators to leverage their idle compute and bandwidth.

The go-livepeer Broadcast node will be implementing an updated selection algorithm that moves the network closer to this vision, and the Livepeer Studio hosted gateway will be updating its B nodes to leverage this change at the beginning of next week.

  1. The canonical selection algorithm in go-livepeer will be updated to (1) use price as a factor in addition to stake, (2) remove the role of pure-play randomness and (3) account for regional performance as an exclusion factor. It will use a multidimensional probability distribution to allocate streams, with tune-able parameters to adjust the relative importance of price and stake. The use of this probability distribution allows broadcasters to preserve some amount of luck, without sacrificing the key goals of optimal price and performance.

    • The Livepeer Studio team will set their initial performance thresholds at 0.65 for Live (slightly better than realtime) and 0.3 for VOD (roundtrip time of 5.7s for a 2s segment)
  2. We will provide a Jupyter notebook as a sandbox environment for Orchestrators and Broadcasters to test the impact of different price and performance characteristics and weights on stream distribution.

These actions are a significant change in the relationship between Broadcasters and Orchestrators; their impact will be closely monitored in the weeks and months to come.


As promised, here is a Jupyter notebook with an implementation of the selection algorithm. This mirrors the implementation in go-livepeer and is provided as a playground to help Os optimize their pricing and performance characteristics.

The link below is view-only, so you’ll need to clone it. I’ll xpost this to Discord as well.

I like the addition of price being incorporated into selection. It makes the open market for transcoding more dynamic.