Uploaded June 2026 | Updated September 2026, 3 weeks ago
NSDI '26 - REAL: Emulating Control Plane at Simulator’s Cost
Ze Xia, Hao Li, Jinyu Fu, Xin Wan, Yihan Dang, and Danfeng Shan, Xi'an Jiaotong University; Li Chen, harnets.ai; Peng Zhang, Xi'an Jiaotong University
Community Award Winner!
Validating control plane behavior and ensuring policy compliance in modern, large-scale networks is a critical challenge. Simulation-based approaches offer low computational and memory costs, but their level of abstraction fails to capture vendor-specific device behaviors, limiting their accuracy for real-world validation. In contrast, control plane emulation provides high fidelity by using unmodified router containers that preserve these vendor-specific details, but its excessive computational and memory requirements make it impractical for large networks. In this paper, we present REAL, a lightweight runtime that emulates control planes using unmodified router containers but at the cost of simulation-based approaches. REAL achieves this by simulating a lightweight data plane to accelerate boot-up, employing a two-phase scheduling policy to minimize cache inefficiencies during convergence, and enabling iterative convergence to reduce peak memory usage by partitioning the network. Our evaluation shows that REAL emulates a 1,000-node network 4× faster than state-of-the-art simulation while preserving vendor-specific behaviors, and can scale to 4,500 nodes on four commodity servers by shaving 8.3× memory.
View the full NSDI '26 program at usenix.org/conference/nsdi26/technical-sessions
NSDI '26 - REAL: Emulating Control Plane at Simulator’s Cost
Ze Xia, Hao Li, Jinyu Fu, Xin Wan, Yihan Dang, and Danfeng Shan, Xi'an Jiaotong University; Li Chen, harnets.ai; Peng Zhang, Xi'an Jiaotong University
Community Award Winner!
Validating control plane behavior and ensuring policy compliance in modern, large-scale networks is a critical challenge. Simulation-based approaches offer low computational and memory costs, but their level of abstraction fails to capture vendor-specific device behaviors, limiting their accuracy for real-world validation. In contrast, control plane emulation provides high fidelity by using unmodified router containers that preserve these vendor-specific details, but its excessive computational and memory requirements make it impractical for large networks. In this paper, we present REAL, a lightweight runtime that emulates control planes using unmodified router containers but at the cost of simulation-based approaches. REAL achieves this by simulating a lightweight data plane to accelerate boot-up, employing a two-phase scheduling policy to minimize cache inefficiencies during convergence, and enabling iterative convergence to reduce peak memory usage by partitioning the network. Our evaluation shows that REAL emulates a 1,000-node network 4× faster than state-of-the-art simulation while preserving vendor-specific behaviors, and can scale to 4,500 nodes on four commodity servers by shaving 8.3× memory.
View the full NSDI '26 program at usenix.org/conference/nsdi26/technical-sessions
