Spot Instances
Definition
Spot Instances are spare cloud computing capacity available at steep discounts (up to 90% off), but can be reclaimed by the provider with short notice.
Use Cases
- Pinterest: Large-scale batch processing and data workloads with cost optimization — Used Amazon EC2 Spot Instances as part of their compute fleet for fault-tolerant, horizontally scalable jobs, designing workloads to tolerate interruptions and re-queue work when instances are reclaimed. (Reduced compute costs for suitable batch workloads while maintaining throughput by building interruption-tolerant processing.)
- Skyscanner: Cost-efficient CI/CD and build/test workloads — Ran elastic build and test capacity on Amazon EC2 Spot Instances, scaling out for peak demand and relying on automation to replace interrupted capacity. (Lowered infrastructure spend for bursty, parallelizable workloads while keeping developer pipelines responsive through autoscaling.)
- Netflix: Cost-optimized, scalable compute for fault-tolerant services and batch-style processing — Adopted Amazon EC2 Spot Instances within an automated, resilient compute platform that replaces interrupted instances and spreads capacity across instance types and Availability Zones. (Improved cost efficiency at scale by using discounted capacity while maintaining reliability through automation and redundancy.)
Provider Equivalents
- AWS: Amazon EC2 Spot Instances
- Azure: Azure Spot Virtual Machines
- GCP: Google Cloud Spot VMs
- OCI: OCI Preemptible Instances
Frequently Asked Questions
- What's the difference between Spot Instances and Reserved Instances (or Savings Plans)?
- Spot Instances are discounted but can be interrupted with short notice when the cloud provider needs the capacity back. Reserved Instances and Savings Plans provide discounts in exchange for a commitment (time and/or spend) and are not interruptible. Use Spot for flexible, fault-tolerant work; use reservations/commitments for steady, always-on workloads.
- When should I use Spot Instances?
- Use Spot Instances for workloads that can handle interruptions: batch processing, CI/CD runners, big data jobs, containerized workers, media transcoding, Monte Carlo simulations, and machine learning training with checkpointing. Avoid Spot for single-instance stateful systems, critical low-latency services without redundancy, or workloads that can’t restart safely.
- How much does Spot Instances cost?
- Spot pricing is typically much lower than on-demand (often 50–90% less), but the exact price and availability vary by region, instance type, and current supply/demand. You pay for the time the instance runs, and the main trade-off is the risk of interruption. Your total cost depends on how often you’re interrupted and how efficiently your workload resumes.
Category: cloud
Difficulty: intermediate
Related Terms
See Also