Availability Bandwidth Allocation Method in Inter Data Center Wide Area Network

Shared Inter Data Center Wide Area Networks (inter-DC WANs) increasingly carry both delay-sensitive ISPfacing traffic and bulk inter-DC transfers. Existing traffic engineering (TE) often optimizes latency or utilization in isolation, while availability guarantees are typically coarse-grained and weakly tied to SLA outcomes. We present Tri-classification Traffic Engineering (TTE), an availability-aware TE framework for shared inter-DC WANs that differentiates three traffic classes with heterogeneous latency and bandwidth-availability targets, and jointly optimizes egress selection and multi-path routing under an SLA-driven objective. TTE formulates the control problem as a constrained optimization that couples maximum link utilization, end-to-end latency, and probabilistic bandwidthavailability requirements. To enable fast adaptation to demand fluctuations, TTE learns a deep reinforcement learning agent to recommend class-aware egress decisions, while a lightweight optimizer computes feasible routing splits and availability allocations. Trace-driven experiments on WAN topologies demonstrate that TTE improves SLA satisfaction and latency-utilization trade-offs with low decision latency.