Switchboard V2 (pt.2) — Architecture and Data Flows

In a previous article we looked at the different stakeholders and how they operate within Switchboard V2. Today’s article will focus on the technical components and how they interact with each other. So let’s jump right in.


What is a Data Feed?

A data feed is the centerpiece of Switchboard and is what on-chain developers will use when building smart contracts. A data feed is a collection of jobs that get aggregated to produce some deterministic result. Each job is associated with an endpoint and has a number of tasks that get executed in sequential order in order to produce a single value. Typically the first task in a job will fetch external data with subsequent tasks responsible for parsing the response and transforming the value into a single data type, like an integer or decimal. When an oracle is assigned to process a data feed update, the oracle executes the defined jobs and publishes the median result on-chain. The data feed then computes the final value as the median response among the assigned oracles. In summary, the data feed is the blueprint for how data gets fetched from off-chain sources.

Along with the jobs, a data feed also includes a configuration dictating how often a feed should be updated and the minimum number of jobs or oracles that must respond before accepting a result. The publisher is ultimately responsible for building a data feed and making the necessary trade-offs as it’s a careful balance between cost and update interval. The publisher is usually the on-chain consumer of the data and will have the most familiarity with how the data may be used to make these considerations.

Types of Data Feeds

Once a data feed has been configured, it needs to be assigned an oracle queue to process updates. Data feeds can be public, where they are approved by the DAO and have access to its oracle queue, or private, where the publisher has their own oracle infrastructure or agreements with oracle operators to process their updates. A private feed has the added benefit of embedding API keys within a job for any endpoints that require authentication, meaning a greater level of trust is needed between publishers and oracles. These keys will need to be created by individual oracle operators or provided by the private feed creator. The rest of this article will be focused on public feeds requiring DAO approval.

Lease Contract

Oracle queues have a finite amount of resources proportional to its number of oracles so the DAO may reject new feeds that could cause delays on existing feeds. The publisher is responsible for creating a lease contract to reserve a set amount of computer power from the oracle queue. Once a publisher’s feed is accepted by the DAO, the feed is added to the network and granted permissions to use the oracle queue resources.

Upon creating a lease contract, the publisher may specify a withdrawal authority and fund the lease contract to reward oracle operators for processing any future updates. The withdrawal authority allows a publisher to cancel and refund their lease contract at any moment and specifies where any remaining lease balance is sent. The lease fee is derived from the oracle rewards dictated by each oracle queue. This value can be increased over time to entice additional oracles to join the queue or decreased to entice additional publishers to submit new feeds. If a data feed’s lease is low on funds, any update requests will fail and the feed will be removed from any scheduled updates. As a feed gains popularity, other dApp developers will be incentivized to extend the lease and keep it active. This creates a natural decay where unused feeds go unfunded to make room for new use cases.

Oracle Queue

The Switchboard DAO governs how its oracle resources get allocated and rewarded. Oracles are arranged in a round-robin fashion, where once requested, the next N oracles in the queue are assigned to a feed and cycled to the bottom. Oracle positions are periodically swapped to mitigate oracles being assigned to the same feeds each cycle. A single Oracle queue was architected to support over 100,000 oracles, but given rent costs, the initial implementation will cap queue sizes at 2048 and increase it as the network grows.


The Crank

Data feeds can also be configured so they are updated periodically or on-command, depending on the feed’s use cases. Solana has no mechanism to schedule periodic updates so a Crank is used to jump start the system. Any feed approved by the DAO is free to join the oracle queue’s crank. The Crank is a priority queue of data feed public keys ordered by the feeds next available update time. When cranked, Switchboard will look for any data feeds ready for an update and if successful, reward the user who called it. If no data feed is ready to be updated, the crankers transaction will fail and they could potentially lose their transaction fee. The crank is the scheduler behind the oracles and incentivizes users to help keep the system spinning. Anyone can compete to turn the crank but there can be only one!

When a data feed update is detected, the oracle queue moves the next N oracles to the back of the queue and passes the data feed public key to each oracle to begin processing the update. Each oracle then reads the data feed configuration, executes each job, then publishes their results on-chain. Oracles are always on the queue and can process multiple feed updates when requested.

Oracle Rewards

If enough oracles successfully respond, an on-chain program will aggregate the assigned oracle results and return the median value as the final result. Each oracle is then scored based on their response. The variance threshold is dictated by the DAO on a per queue basis and is used when determining the validity of an oracle’s response in relation to the accepted, median result. Oracles within the acceptable range are awarded whatever fee is set by the DAO. Oracles who’s responses fall outside the acceptable range will be slashed and must forfeit a set amount of their staked capital. The slashing mechanism disincentivizes oracles from reporting dishonest data and helps protect the system from nodes who may have other incentives to return false data. A future article will detail the various incentives to entice nodes to remain honest.

Each feed keeps track of the number of successful and failed responses. If a feed has persistent failures, then the feed is removed from the oracle queue and the publisher is refunded the remaining balance on the lease contract to the withdrawal authority specified when creating the lease contract.

The diagram below depicts the system in action:

Switchboard V2 gives more ways for you to join the network, whether that’s operating a node, publishing feeds, or cranking the system. We are excited to roll out the next version of our community curated oracles and can’t wait to see what the community builds with it!





Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

The main reason to use C++ over python for your Embedded Systems projects

AWS Transformation Day BENELUX

Beanbo — Your Questions

5 Useful Features of C# .NET Core Lambda function

Creating pixel perfect UI for different screen sizes in Android

Event Pipelines (Part 1): Backpressure and Buffering

Distributed diagnostics microservice pattern

Microservice diagnostics structure

Test Annotations — Part 3

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store



More from Medium

OpenShift Service Mesh on IBM System Z/LinuxONE Part I: Installation

The big red letter `O` with parts of it shifted is the logo of Red Hat OpenShift. The little blue sails are the logo of Maistra, the main component of Red Hat OpenShift Service Mesh. The asymmetrical vesica piscis below the OpenShift logo is the logo of the Kiali subsystem of Service Mesh. The little blue Go Gopher wearing a Tyrolean hat perched atop an IBM LinuxOne chassis is the logo for the Jaeger Tracking system used in Red Hat OpenShift Service Mesh.

Cake processes growing number of bank transactions thanks to data engineering pipeline

Obsidian — Custom Plugins Part 4

A Novice’s experience with Load Testing using Autocannon