dstack vs Koordinator

dstack is an open-source orchestration layer for AI workloads across GPU clouds, Kubernetes, virtual machines, and bare-metal clusters. YAML configurations define fleets, development environments, tasks, services, presets, and volumes, while dstack handles provisioning and job scheduling, including auto-scaling, port forwarding, and ingress. Supported accelerators include NVIDIA, AMD, TPU, and Tenstorrent. Documented backends include AWS, Azure, GCP, Kubernetes, GPU cloud providers, remote SSH hosts, and experimental Slurm. Tasks can use frameworks such as accelerate, torchrun, Ray, and Spark. Users manage resources through the CLI or HTTP API, and the server can run on a laptop or another environment that can access the clusters in use. Services can publish inference endpoints through gateways with HTTPS, custom domains, auto-scaling, and rate limits. The self-hosted open-source stack is free. Server data and project secrets are stored in plaintext by default unless administrators configure AES-256-GCM encryption. TPU support is limited to single-host instances with up to eight cores. GPU Marketplace usage is prepaid, with prices that vary by GPU and provider and appear in the console before provisioning.

Koordinator is a free, self-hosted QoS scheduling system for Kubernetes microservices, AI, big-data, audio, and video workloads. It adds priority and quality-of-service controls so different pods and workload types can share nodes. Application profiling supports resource overcommitment while preserving QoS guarantees, and fine-grained isolation targets latency-sensitive work alongside batch jobs. Customizable policies cover Web Service, Spark, Presto, TensorFlow, and PyTorch. QoSManager tunes pod isolation using profiling, interference detection, and SLO settings. Koord-Queue supports TFJob, PyTorchJob, Spark, Argo Workflow, Ray, and native Kubernetes Jobs, with Kueue AdmissionCheck compatibility. YARN Copilot connects Hadoop YARN jobs with Kubernetes pods. Koordinator works alongside kube-scheduler, using control-plane and node components without invasive native changes. Helm 3.5 or newer and Kubernetes 1.18 or newer are required.