The spatial complexity of real-time interactive video varies significantly within a single session, not only across different content types. This renders static resource allocation inadequate in our scenario: even with oracle knowledge of each session's average content complexity, static methods cannot guarantee acceptable picture quality. Using second-by-second oracle spatial complexity knowledge as a deliberate upper-bound methodology, we quantify the potential of dynamic QoE-aware resource sharing through a utility-based framework. In our scenario of 30 concurrent cloud gaming sessions sharing a bottleneck link, dynamic allocation guarantees acceptable quality (VMAF>=50) for all sessions at 35 Mbps, while no static method achieves this even at 100 Mbps. The framework makes allocation policy explicit and tunable, replacing rigid equalization that degrades all sessions when resources are scarce.

翻译：暂无翻译