• Load‑aware two‑stage scheduling improves P2P download performance across distributed nodes. • Vortex protocol (TLV‑based) cuts large file transfer time by 50% (TCP) and 40% (QUIC). • Request SDK routes user requests via consistent hashing, replacing Kubernetes Service load balancing. • Simplified multi‑cluster deployment uses explicit schedulerClusterID for precise affinity control. • Manager and Scheduler components now feature enhanced performance and optimized resource utilization.

Article Summaries:

  • Dragonfly has released version 2.4.0, adding several key improvements. The new load‑aware scheduling algorithm combines central and node‑level scheduling to boost P2P download efficiency. A Vortex transfer protocol (TLV‑based) replaces gRPC, cutting large‑file download times by up to 50 % and lowering memory usage. An SDK now routes user requests to seed peers via consistent hashing, replacing Kubernetes Service load balancing. Multi‑cluster support is simplified with explicit schedulerClusterID assignment. The release also optimizes Manager and Scheduler resource use, enhances preheating, introduces SHA256‑based task IDs to avoid duplicate downloads, caches HTTP 307 redirects, and deprecates the Go client in favor of a Rust client.

Sources: