How to Compare Interruption Frequencies with EC2 Spot Advisor?

The Amazon EC2 Spot Instance advisor is a free web tool provided by AWS that helps cloud architects and developers evaluate the cost savings and stability of Spot Instances before deploying workloads. By analyzing historical data over the trailing thirty days, the tool displays the average discount compared to On-Demand pricing alongside the expected frequency of interruption for every Spot-eligible EC2 instance type across global regions and operating systems. Using this tool allows you to filter by hardware specifications like vCPUs and memory, sort instance families by their reclamation likelihood, and build resilient, diversified Spot fleets that minimize unplanned downtime while maximizing cloud compute savings.

Understanding Spot Instance Interruption Frequencies

Amazon EC2 Spot Instances let you take advantage of unused EC2 capacity in the AWS cloud at up to a 90% discount compared to On-Demand prices. Because AWS can reclaim this capacity whenever demand for On-Demand instances rises, Spot Instances are subject to interruptions with a two-minute warning.

Rather than providing an exact real-time probability, the Spot Instance advisor groups historical interruption rates over the trailing 30 days into five distinct frequency brackets:

  • Less than 5%: The most stable tier, where less than 1 in 20 instances experience a reclamation event over a month.
  • 5% to 10%: Low interruption frequency, suitable for most fault-tolerant batch jobs and containerized workloads.
  • 10% to 15%: Moderate interruption frequency, requiring reliable checkpointing or fast instance replacement mechanisms.
  • 15% to 20%: Elevated interruption frequency, indicating higher contention for the underlying hardware capacity pool.
  • Greater than 20%: High interruption frequency, best avoided unless your workload is completely stateless and can handle frequent node churn.

Step-by-Step Guide to Comparing Instance Types

Comparing interruption frequencies across instance families takes only a few minutes using the public Spot Instance advisor web interface.

1. Select Your Target AWS Region and OS

Capacity pools are isolated by Region, Availability Zone, and Operating System. An instance type that rarely experiences interruptions in us-east-2 (Ohio) might have a >20% interruption rate in us-east-1 (N. Virginia). Start by selecting your deployment Region and OS (Linux/UNIX or Windows) from the top dropdown menus in the Spot Instance advisor.

2. Filter by Minimum Hardware Requirements

Instead of manually scrolling through hundreds of EC2 instance types, use the resource sliders and filters to narrow down candidates that fit your workload profile:

  • vCPU: Set the minimum and maximum virtual CPU cores required per node.
  • Memory (GiB): Define the RAM boundaries for your application.
  • Instance Type / Family: Use the search bar to filter by specific families (such as m5, c6i, or r6g) or processor architectures (such as AWS Graviton, Intel, or AMD).

3. Sort by Frequency of Interruption

Click the Frequency of interruption column header in the comparison table to sort the filtered results in ascending order. This immediately brings the most stable instance types (<5% interruption rate) to the top of your view, paired alongside their Savings over On-Demand percentage.

4. Compare Generations and Architectures

Look across different generations and CPU architectures that meet the same vCPU and memory specifications. You will often find that:

  • Previous-generation instances (e.g., m5 or c5) and newest-generation instances (e.g., m7i or c7g) have different demand profiles depending on regional adoption.
  • Graviton-based instances (such as m6g, c6g, and r6g) frequently offer both lower interruption frequencies and higher percentage savings compared to equivalent x86 instances.
  • Different sizes within the same family vary significantly. For example, an m5.2xlarge might sit in the 10–15% bracket, while two m5.xlarge instances sit in the <5% bracket.

Calculating Expected Fleet Availability and Cost

When designing a multi-instance pool, you can model expected cost and interruption exposure mathematically. If an instance type has an On-Demand hourly price \(P_{OD}\) and a Spot discount fraction \(D\), the effective hourly Spot price \(P_{Spot}\) is:

\[P_{Spot} = P_{OD} \times (1 - D)\]

Furthermore, if you distribute stateless workloads across \(N\) independent capacity pools where each pool \(i\) has an independent interruption probability \(p_i\) over a given window, the probability \(P_{\text{all}}\) that all pools experience simultaneous interruptions is reduced exponentially:

\[P_{\text{all}} = \prod_{i=1}^{N} p_i\]

For example, combining \(N = 3\) distinct instance pools that each fall into the \(<5\%\) bracket (\(p_i \le 0.05\)) lowers the theoretical likelihood of simultaneous pool exhaustion to at most \(0.05^3 = 0.000125\), or \(0.0125\%\).

Best Practices for Acting on Spot Advisor Data

  • Diversify Across Multiple Instance Families: Never rely on a single instance type based on Spot Instance advisor data alone. Select at least 5 to 10 compatible instance types in the <5% and 5–10% brackets to include in your EC2 Auto Scaling group or Amazon EKS node group.
  • Use the price-capacity-optimized Allocation Strategy: While the Spot Instance advisor shows trailing 30-day regional averages, EC2 Auto Scaling and Karpenter can use real-time capacity data at launch time. Pairing your advisor-curated instance list with the price-capacity-optimized strategy ensures AWS automatically provisions from the deepest, least-interrupted Availability Zone pool available at that exact moment.
  • Re-evaluate Periodically: Spot capacity supply and On-Demand demand shift over time due to seasonal traffic and new hardware rollouts. Review the Spot Instance advisor quarterly to refresh the instance types defined in your launch templates.