cedana

What

Cedana is a compute orchestration platform for GPU and CPU workloads. It is designed for teams running AI inference, AI training, agents, gaming infrastructure, and HPC workloads that need better throughput, lower interruption risk, and more flexible use of on-premise and multi-cloud infrastructure.

The product extends existing orchestration environments such as Kubernetes and SLURM rather than replacing them. Based on the page, its core workflow is to schedule, checkpoint, migrate, resume, and fail over stateful workloads in real time according to price, performance, SLAs, and resource availability, with a strong focus on reliability and utilization.

Features

• Real-time workload scheduling and migration: Cedana matches workloads to available resources based on price, performance, SLAs, and capacity to improve throughput and responsiveness.
• System-level checkpointing and restore: It continuously saves workload state so jobs can resume after GPU or CPU failures without starting over.
• Support for stateful workload failover: Automatic failover helps preserve progress for long-running and mission-critical jobs such as training, inference, and agents.
• Extension of existing orchestrators: The platform is described as working with Kubernetes, Kueue, KServe, Kubeflow, SLURM, and Ray, which helps teams adopt it within current environments.
• Elastic scaling and downscaling: Cedana can scale workloads and clusters up or down, including preempting and saving workloads so resources can be reduced without losing progress.
• Live migration and dynamic resizing: The site highlights GPU live migrations and resizing workloads onto more suitable instances without interruption, which can improve utilization and placement efficiency.

Helpful Tips

• Verify fit by workload type: Cedana appears most relevant for stateful, long-running, or interruption-sensitive compute jobs where checkpointing and migration deliver clear operational value.
• Assess orchestration maturity first: Organizations already using Kubernetes, SLURM, or adjacent ML/HPC tooling are likely to have a faster path to evaluation because Cedana is positioned as an extension layer.
• Validate claims in a controlled environment: The site presents performance and utilization improvements, but buyers should confirm expected gains against their own workload mix, failure patterns, and infrastructure topology.
• Map adoption to operational pain points: The strongest use cases seem to be spot usage, failover, zero-downtime upgrades, and dynamic resizing, so prioritization should start with the most expensive or failure-prone workflows.
• Review checkpointing behavior carefully: For distributed training and inference systems, implementation teams should examine checkpoint frequency, resume behavior, and operational overhead in their specific stack.

OpenClaw Skills

Cedana could likely pair well with OpenClaw in infrastructure operations, AI platform engineering, and workload governance workflows. A likely use case would be OpenClaw agents that monitor queue depth, SLA risk, spot market conditions, and cluster health, then trigger Cedana-based migration or scaling policies through documented APIs and orchestration layers. The site does not confirm a native OpenClaw integration, so this should be treated as a workflow design opportunity rather than a built-in capability.

In practice, OpenClaw skills could be built for capacity planning, failure-response automation, cost-aware job placement, and workload-specific runbooks across training, inference, and HPC environments. That combination could shift platform teams from manual cluster operations toward policy-driven compute management, with OpenClaw handling decision logic and operator workflows while Cedana handles stateful checkpointing, migration, and workload continuity.