v2.4
Manager
- Optimize memory and CPU usage.
- Regularly clean up inactive schedulers and seed peers.
- Support specifying scheduler cluster ID in client configuration, allowing the client to connect to a specific scheduler cluster.
Scheduler
- Implement scheduler-aware peer bandwidth scheduling.
- Optimize the scheduling algorithm to improve bandwidth utilization in the P2P network.
Client
- Support RDMA for faster network transmission in the P2P network, enhancing the loading of AI inference models into memory.
- Support TCP and QUIC protocols for downloading files, allowing users to choose the protocol that best suits their needs.
- Optimize file transfer speed in the P2P network.
- Encrypted storage of downloaded content offers a more secure solution.
- Support selecting the optimal parent for downloading based on QoS (Quality of Service) metrics.
Others
- Add auto injection webhook and integration with helm chart for easier deployment and management.
- Add more E2E tests and unit tests.
Documentation
- Restructure the documentation to make it easier for users to navigate.
- Enhance the landing page UI.
- Add more documentation on system interactions and implementation details.
AI Infrastructure
- Optimize large file distribution within the infrastructure.
- Optimize handling of a large number of small I/Os for Nydus.