This week, we showcased groundbreaking advancements across in-network AI, GPGPU virtualization, edge storage, and post-quantum cryptography (PQC):
🔹 In-Network AI: Distilled DNNs into P4 switches using cascaded LUTs, achieving 93% F1-score for DDoS detection and 96% less memory usage—enabling wirespeed inference without hardware changes.
🔹 GPGPU Virtualization: RDMA outperformed TCP/IP in GPGPU remoting, slashing Matrix Multiplication time by 5.5x and reducing context switches by 98% with GVirtuS.
🔹 Edge Storage: CATER optimized Apache Ozone for constraint-aware placement, cutting active nodes by 23% while meeting all energy, latency, and compliance goals.
🔹 Quantum-Resilient Networking: Benchmarked Falcon, Dilithium, and Kyber on NVIDIA DPUs, achieving 100 Gb/s post-quantum secure key exchange. Falcon excels in lightweight edge environments, while Dilithium/Kyber dominates in data centers.
🔹 Hybrid PQC Stacks: Combined Kyber’s KEM with Falcon/Dilithium signatures for scalable, quantum-resistant infrastructure—from edge to cloud.
🔹 Confidential Computing: Embedded PQC into DPUs, securing data-in-use and data-in-transfer across the edge-cloud continuum.
📚 Explore the full papers and updates: