Sequential solver-free traffic engineering for data centers

The authors present a new Sequential Source-Destination Optimization (SSDO) algorithm as a fast, solver-free method for traffic engineering in large-scale data center networks, evaluated on Meta DCNs and wide-area networks.

• SSDO decomposes traffic engineering into sequential SD subproblems, optimizes per-demand split ratios using a Balanced Binary Search Method (BBSM) to minimize Maximum Link Utilization (MLU), and dynamically reorders SDs based on real-time link utilization to speed convergence and improve solution quality.
• In a Meta data center topology, SSDO achieves 65% and 60% lower normalized MLU versus TEAL and POP respectively, and delivers a 12× speedup over POP; the work is accepted to NSDI ‘26, to be held in Seattle, USA, May 4–6, 2026.