Frequently Asked Questions
Find answers to common questions about Compute Gardener's sustainable computing solutions.
Scheduler
What types of workloads work best with the scheduler?
The scheduler is most effective with batch processing jobs, ML training workloads, and other non-time-critical operations that can be flexibly scheduled. It's particularly valuable for compute-intensive tasks that can be shifted to times of lower carbon intensity or electricity costs.
How does the scheduler determine when to run workloads?
The scheduler uses real-time carbon intensity data from Electricity Map API and configurable time-of-use pricing schedules. It evaluates these signals against customizable thresholds, considering both carbon emissions and cost optimization goals. The scheduler also tracks pod metrics to estimate power consumption and carbon emissions, with configurable sampling intervals and downsampling strategies, providing detailed insights into workload energy impact.
Can I override scheduling decisions for specific workloads?
Yes, you can use pod annotations to opt out of carbon-aware or price-aware scheduling, or set custom thresholds for specific workloads. You can also set custom maximum scheduling delays using formats like "12h", "30m", or "1h30m" through annotations. For advanced use cases, you can enable pod priority-based scheduling to ensure critical workloads are scheduled promptly. This gives you fine-grained control over how each workload is scheduled.
Integration & Setup
What are the prerequisites for installation?
The critical prerequisite is having metrics-server installed and running in your Kubernetes cluster, as it's required to collect CPU metrics for power estimation. Without metrics-server, power-related metrics will report as 0, impacting the scheduler's ability to make optimal decisions. You will also need an Electricity Map API key for carbon intensity data. For monitoring, the scheduler can integrate with Prometheus through included ServiceMonitor resources and annotations. The scheduler is designed to work with standard Kubernetes installations and can be deployed using Helm (recommended) or standard YAML manifests.
How does the scheduler handle API rate limits?
The scheduler includes built-in caching of API responses to limit external API calls and respect rate limits. The cache duration is configurable, allowing you to balance between data freshness and API usage. This ensures efficient operation while maintaining real-time decision-making capabilities.
How does it integrate with existing infrastructure?
The scheduler integrates as a Kubernetes scheduler plugin, working alongside your existing scheduling policies. It supports various infrastructure setups including on-premise, cloud providers (AWS, GCP, Azure), and hybrid environments.
What monitoring and observability features are available?
The scheduler exports Prometheus metrics for carbon intensity, electricity rates, scheduling decisions, and estimated savings. It also provides detailed metrics for CPU usage, memory usage, GPU utilization, power consumption estimates, metrics cache statistics, job energy usage, and carbon emissions. The scheduler includes configurable metrics sampling intervals, downsampling strategies, and retention periods for completed pods. These metrics can be integrated with your existing monitoring stack for comprehensive observability.
Enterprise API
What additional features does the Enterprise API provide?
The Enterprise API offers advanced real-time optimization signals, combining carbon intensity, spot instance pricing, and electricity costs. It includes sophisticated scheduling algorithms, enhanced analytics, and priority support.
How is Enterprise API pricing structured?
Enterprise API pricing is usage-based with volume discounts. We work with each organization to understand their needs and provide custom pricing plans that align with their infrastructure scale and optimization goals.
Carbon Credits Program
How are carbon savings calculated and verified?
Carbon savings are calculated based on workload power consumption and the difference in carbon intensity between actual execution time and baseline scenarios. Our methodology is being developed with carbon market experts and will undergo independent verification.
When will carbon credits be available?
The carbon credits program is scheduled to launch in late 2025. Early adopters of our scheduling technology will have priority access to the program and can influence its development through our pilot phase.
Still Have Questions?
Contact our team for answers to your specific questions or to discuss your sustainable computing needs.