Data Center Transmission Control Protocol (DCTCP) is an excellent transmission control protocol for data centre network (DCN). Although DCTCP can achieve the high throughput and stability, it trades off the convergence time. It takes a long time for a new flow to increase its congestion window ( cwnd ), which greatly increases the completion time of some delay sensitive short flows, such as query traffic, short messages existing extensively in DCN. An improved DCTCP (IDCTCP) is proposed to improve the convergence of DCTCP. IDCTCP utilises the receive window field in TCP header to communicate sender's current cwnd to a switch. The switch only marks the packets with large cwnd . So the flows with small cwnd are protected and can increase their cwnd quickly, which can effectively improve the convergence time of DCTCP. As a result, the short flows completion time (SFCT) can be reduced. The performance of IDCTCP has been validated by NS2 simulations.

Artificial intelligence is reshaping education, yet research remains fragmented across technological streams and often conflates technical capability with demonstrated educational effectiveness. This narrative review synthesizes representative literature published mainly between 2010 and 2026 to trace the evolution of AI-enabled education from computer-assisted instruction toward adaptive, immersive, generative, and potentially more general forms of intelligence. It compares three major technological streams: intelligent tutoring and adaptive learning systems, extended reality (XR), and large language model-based generative AI. The review finds that intelligent tutoring systems support learner modeling, adaptive feedback, and personalization, but their evidence base is methodologically heterogeneous. XR shows comparatively more consolidated evidence, particularly through systematic reviews and meta-analyses, with generally positive but context-dependent effects on engagement and learning outcomes. Generative AI expands dialogic explanation, scaffolding, content generation, and personalized support, although learner-outcome evidence remains limited relative to capability benchmarking. Building on these findings, the review proposes a Technology–Mechanism–Outcome–Challenge framework and argues that future progress toward artificial general intelligence (AGI) should be evaluated through learning mechanisms, measurable outcomes, human oversight, equity, and responsible governance.