Node Competition-Task Acquisition
Taking LLM training or data crawling as examples, these types of nodes are usually for research-oriented businesses, such as cancer and earthquake prediction, which require continuous and significant computing power for competition. To avoid severe resource waste, we have set stricter requirements for the node's owner. The challenge of this business is to classify Tasks, allowing for distributed computing of multiple agents.
The guardian nodes classifies and orders tasks, as shown in the following figure, where tasks in a certain round are allocated to several Task queues, and each Task queue has an order list for counting nodes.
As shown in the figure below: Nodes 1-6 have chosen Task1. The order list counts the Nodes that have chosen this Task.
Modes 7-11, upon reviewing all Task queues, finds that Task 1 is extremely competitive, so they proactively choose Task 2 and update the order list 2.
At this moment, the newly joined Node X comprehensively analyzes its own computing power and the difficulty of the Task, proactively choosing Task X because at this time, Task-X has the lowest difficulty and no competition.
Six Nodes that chose Task 1 have returned their computational results, but the Agetn-X's Monitor process will only select the first result submitted. However, to balance the distribution of rewards, Task 1 can also choose to retain the results provided by the top three nodes, similar to Ethereum's uncle block incentive, but this increases the complexity of the incentive, and we will address this issue in more detail in the future.
Last updated