Research & Publications

Research & Publications

Explore research from NgKore Foundation members advancing open-source telecom, quantum security, and next-generation networking.

conference paper

Saving 5G Core from Cryptocrastination

Aditya KorangaOctober 2025ACM free5GC '25, Taipei, Taiwan

A comprehensive post-quantum cryptography implementation across all 5G core network functions and interfaces. Demonstrates ML-KEM achieving over 4 million operations per second on 80-core systems, with GPU acceleration reaching 20 million ops/sec. Shows practical migration pathways with only 27-35% overhead for critical operations.

PQC5G CoreML-KEMML-DSAQuantum Security
Download PDF
conference paper

Accelerating AI Microservice Communication with eBPF Shared Sockets: A Kernel-Level Framework for Low-Latency Agentic Workloads

Satyam Dubey, Nikhil Kumar RajputMay 2026University of Delhi

A shared socket acceleration framework based on eBPF that intercepts local traffic at the socket layer, bypassing TCP segmentation, checksum calculations, and netfilter hooks entirely. Uses SOCKHASH maps with sockops and sk_msg programs to directly inject payloads into receiver buffers for low-latency agentic AI workloads.

eBPFAI MicroservicesLLMShared Sockets
Download PDF
technical report

eBPF-Powered Cross-Layer Anomaly Detection: A Unified Kernel-Space Security Monitoring Framework for Linux Systems

Nikhil Jangid, Nikhil Kumar RajputMay 2026University of Delhi

A unified kernel-space security monitoring framework that instruments the Linux kernel at three vantage points: XDP hook, TC classifiers, and syscall tracepoints. Achieves 1.97 microsecond per-packet processing at XDP and 4.99 microseconds at TC, extracting 47 cross-layer features for anomaly detection.

eBPFXDPSecurityAnomaly DetectionLinux Kernel
Download PDF
conference paper

Kernel-Level GPU Observability for LLM Workloads via eBPF and the Linux DRM Subsystem

Khushi, Nikhil Kumar RajputMarch 2026IEEE Transactions on Parallel and Distributed Systems

Introduces DRMMON, a two-layer GPU monitoring framework that hooks eBPF programs into amdgpu and dma_fence tracepoints for job-by-job scheduling visibility. Profiled Mistral-7B and LLaMA-2-7B during fine-tuning and inference with under 1.5% runtime overhead on AMD Instinct MI300X and Intel Arc systems.

eBPFGPULLMDRMObservability
Download PDF

Have research to share?

We welcome contributions from researchers and engineers working on open-source telecom technologies.

Get in Touch