Tensor Fusion is a state-of-the-art GPU virtualization and pooling solution designed to optimize GPU cluster utilization to its fullest potential.
-
Updated
Nov 19, 2026 - Go
Tensor Fusion is a state-of-the-art GPU virtualization and pooling solution designed to optimize GPU cluster utilization to its fullest potential.
Kubernetes DRA driver for exclusive, topology-aware CPU allocation for high-performance workloads
Slurm on Kubernetes, enabling workloads from both Slurm and Kubernetes with Kueue-based scheduling.
Kubernetes DRA driver for deterministic CPU time-sharing using Linux SCHED_DEADLINE. Partitions cores into guaranteed, non-overlapping time slots exposed as DRA devices.
🦄 Distributed Inference on Kubernetes with DRA and MIG
Linux + Kubernetes
Developed as a semester project for CS-315 (Cloud Computing), this project implements a robust, end-to-end data processing pipeline designed for cloud environments. It focuses on the core cloud principle of elasticity by implementing a system that dynamically adjusts computing resources (Spark Executors) based on real-time traffic volume.
⚡ Cloud-Native AI/ML & Jupyter Lab Platform powered by Kubernetes Operator & Dynamic Resource Allocation (DRA). Automates GPU scheduling, stateful workspaces, and multi-framework ML lab environments.
Provider-neutral Kubernetes DRA observability, simulator, and diagnostics for GPU and accelerator workloads—model virtual device pools, inspect ResourceClaim/ResourceSlice state, and explain allocation outcomes without physical hardware.
Kubernetes Network (DRA) Driver orchestrating host-backed networks across physical and virtual network interfaces (Macvlan, IPVLAN, SR-IOV VF, RDMA...)
Crop-disease inference on Kubernetes: a three-tier ladder, GPU scheduling, KEDA autoscaling on queue depth rather than CPU, and warm headroom for a cold start you cannot scale away. M1 and M2 built and measured; the on-device tier runs in the browser. The measurement says the ladder fails on field photos, and the README says so.
Anchor is a Kubernetes DRA driver that gives pods static AWS IP addresses which follow them when they are rescheduled. most common use case is SCTP/SIGTRAN in EKS and on-premises
Kubernetes DRA driver for deterministic L3 cache partitioning via resctrl (Intel RDT, AMD QoS, ARM MPAM). Partitions cache ways into non-overlapping slices exposed as DRA devices.
Kubernetes x NVIDIA DRA Workshop
Workload-aware GPU scheduling for Kubernetes DRA: grades each GPU per workload, keeps degraded GPUs out of training and serving inference. Measured on real NVIDIA GPUs.
Kubernetes DRA driver allocating AF_XDP queues and HSR/PRP redundant streams (IEC 62439-3) to pods
high-performance, resilient Python framework for building Kubernetes Dynamic Resource Allocation (DRA) hardware drivers.
🔬 6G Network Slicing with AI-Driven Dynamic Resource Management | Multi-Agent DRL (SAC) | Explainable AI (SHAP) | Energy-Aware Optimization | Dynamic Soft Slicing | MLOps Pipeline | FastAPI | SimPy | PyTorch
Explain why Kubernetes can't allocate a GPU to your pod: a kubectl plugin that replays the scheduler's DRA allocator
A library and a cli to view K8s cluster resources
To associate your repository with the dynamic-resource-allocation topic, visit your repo's landing page and select "manage topics."