v2.3

Manager:

• Configure scheduling weights.
• Support scopes for Personal Access Tokens (PATs).
• Regularly clean up inactive schedulers and seed peers.
• Display more Peer information in the console, such as CPU and memory usage.
• Display persistent cache information of peers in the console.
• Add management of sync peers in the console.

Scheduler:

• Optimize the scheduling algorithm to improve bandwidth utilization in the P2P network.

Client:

• Support RDMA/QUIC for faster network transmission in the P2P network, enhancing the loading of AI inference models into memory.
• Define a codable protocol for data transmission, providing faster encoding/decoding.
• Support persistent cache, allowing access within the P2P cluster without uploading to other storage, facilitating faster read/write of AI models and datasets.
• Allow peers to get the QoS of parents and select the optimal parents for downloading.
• Preheat files in the memory cache to improve download speed.

Others:

• Add more performance tests in the dfbench command.
• Add more E2E tests and unit tests.

Documentation:

• Restructure the documentation to make it easier for users to navigate.
• Enhance the landing page UI.

AI Infrastructure:

• Optimize large file distribution within the infrastructure.
• Optimize handling of a large number of small I/Os for Nydus.