In December we made improvements to AutoAgora and continued to study its performance and its impact on the query market. We also made progress on Verifiable Scalar. We began setting up a Gateway that we will operate. And we partnered with the liquid staking protocol Tenderize.me.
We continued our investigation of the impact of AutoAgora, our tool for automated query pricing for Indexers. This task included analysis of different metrics used in dynamic pricing: revenue, regret, and query volume ratio. We accessed AutoAgora logs and information derived from the Gateway and compared the performance of the Semiotic Indexer running AutoAgora with other Indexers that are not using AutoAgora.
Our initial proposal for using H2S2 for Scalar, presented at the Graph R&D retreat, made an invalid assumption that receipts could be securely re-used. We discovered that receipts cannot be securely reused in H2S2 (and Homomorphic Signature Schemes in general) and so have to modify our Scalar design.
Without receipt re-use, one of the restrictions with H2S2 (in the context of Scalar) is that it assumes a maximum table size and that an Indexer has collected enough receipts to fill this max table size, if the Indexer has less than the max number then they cannot collect any of the value they are owed. We proposed a proving system, which can be used to augment H2S2 and allows the Indexer to collect the value they are owed even if they have less than the max number of receipts. We’re calling this a "Well-Formedness Proof” (WFP). We have a draft paper describing H2S2 augmented with WFP (H2S2+WFP) and have circulated it internally, i.e. with the Semiotic Labs and Edge & Node cryptography teams.
H2S2+WFP still has some restrictions in the context of Scalar. Specifically, the Gateway must somehow ensure that no receipt identifier is repeated (the receipt reuse problem above). This impacts the architecture of the Gateway and might also limit the max number of receipts that can be claimed using H2S2+WFP. We spent December working on a proposal for a new protocol leveraging H2S2+WFP and combining it with
rowID and signing keys management that will fulfill the throughput requirements of Scalar.
One of the Foundation’s priorities for 2023 is to open up the Gateway and enable other core devs to run trusted Gateways. We predict that understanding the Gateway’s architecture and all the nuances associated with running a Gateway will be a multi-month effort. In December we studied several Gateway-related documents provided by E&N. We began to explore E&N’s implementation. In our investigation we have mostly focused on two Gateway components:
- The Indexer Selection Algorithm (as it ties to our work on AutoAgora and our ongoing investigation of its impact in production), and
- Studio (as we are trying to understand how Studio can be used for a tiered billing model).
We refreshed our Indexer infrastructure on GCP with more infrastructure-as-code (basically a lot of work with Terraform). We also worked on node infrastructure on bare metal leveraging Kubernetes for running Ethereum nodes.
We will continue to investigate AutoAgora’s performance (and refine the yellow paper as necessary). We also will implement a metrics module (with metrics we applied to production data) and incorporate it into our training simulator. Finally, we will improve AutoAgora based on the feedback that we have received from Indexers.
We will continue work on the protocol described above, with the goal of having a proof of concept at the end of the month. We will also work on the paper describing H2S2+WFP with the goal of sharing it with a broader community for security review.
Moving into the new year, we want to ensure that we align our priorities with those of the Foundation. In that light, we will focus on standing up our own Gateway, even if we just run E&N’s code. We do not presently have a confident idea of how much effort that will require. It may well be a multi-quarter effort.
Our Indexer automation tooling also includes the Allocation Optimiser, on which we worked in collaboration with E&N and GraphOps. We want to improve the quality of that software, potentially moving to a compiled binary. We also have a yellow paper in the works for the Allocation Optimizer. Until we release it, feel free to look at the first of our blog posts on the subject.
We also support the protocol economics group (PECON). Last year, PECON ended with some interesting discussions around curation, Indexer rebates/slashing, and indexing rewards. A few of us in the group have also started early discussions around Warpsync economics, so that will be on the horizon as well. We will continue to engage in these discussions going forward.