Spot Fleet Strategies: Lowest Price, Diversified, Capacity Optimized

Spot Fleet is a service that simplifies launching and managing a fleet of Spot Instances. Spot Instances are spare EC2 capacity that AWS offers at up to 90% discount compared to On-Demand, but they can be reclaimed with a 2-minute warning when AWS needs the capacity back. Spot Fleet aggregates multiple Spot Instance requests across instance types and Availability Zones to achieve desired capacity, performance, and cost. The strategies determine how Spot Fleet selects which instance pools (a combination of instance type and Availability Zone) to launch from.

When you create a Spot Fleet, you define a target capacity (number of instances or vCPUs/memory) and a launch specification that includes one or more instance types and subnets (AZs). Spot Fleet then monitors the Spot market and launches instances from the pools that best meet the chosen strategy. The fleet continually evaluates the market and can replace instances that are reclaimed (interrupted) with new ones from different pools, maintaining the target capacity.

#### 1. lowestPrice (Default)

This strategy launches instances from the pool that has the lowest Spot price at the time of launch. It is the simplest and most cost-focused. How it works:

Use case: When you want absolute lowest cost and can tolerate the risk of running in a single AZ/instance type, or if you are running batch jobs that can be interrupted and restarted.

Risk: If the cheapest pool becomes unavailable (e.g., price spikes or capacity disappears), the fleet may not be able to maintain capacity. Single-AZ deployment reduces resilience.

#### 2. diversified

This strategy distributes Spot Instance launches across multiple pools. How it works:

Use case: When you need high availability and want to avoid putting all eggs in one basket, but still want to optimize cost by preferring cheaper pools. Good for stateless web servers or microservices.

Risk: Slightly higher cost than lowestPrice because you may use more expensive pools. But much more resilient.

#### 3. capacityOptimized

This strategy launches instances from the pool that has the most available Spot capacity. How it works:

Use case: When you need to launch a large number of instances quickly and cannot tolerate interruptions, such as for urgent data processing, rendering, or large-scale simulations.

Risk: Higher cost because you may use expensive pools. But maximizes the chance of getting and keeping capacity.

To create a Spot Fleet with the diversified strategy using the AWS CLI:

aws ec2 request-spot-fleet \
    --spot-fleet-request-config file://config.json

Example config.json:

{
    "TargetCapacity": 100,
    "AllocationStrategy": "diversified",
    "SpotPrice": "0.50",
    "IamFleetRole": "arn:aws:iam::123456789012:role/aws-ec2-spot-fleet-tagging-role",
    "LaunchSpecifications": [
        {
            "InstanceType": "c5.large",
            "ImageId": "ami-0abcdef1234567890",
            "SubnetId": "subnet-0a1b2c3d4e5f67890"
        },
        {
            "InstanceType": "m5.large",
            "ImageId": "ami-0abcdef1234567890",
            "SubnetId": "subnet-0a1b2c3d4e5f67891"
        }
    ],
    "TerminateInstancesWithExpiration": true
}

To verify the fleet:

aws ec2 describe-spot-fleet-requests
aws ec2 describe-spot-fleet-instances --spot-fleet-request-id sfr-1234567890abcdef0

The exam tests your ability to choose the right strategy based on requirements: - Cost optimization with tolerance for single-AZ failure? → lowestPrice - High availability and cost savings? → diversified - Large capacity quickly, cost is secondary? → capacityOptimized

Remember: lowestPrice uses ONE pool, diversified uses MULTIPLE pools, capacityOptimized uses ONE pool but picks the one with most capacity.

A pharmaceutical company runs drug discovery simulations that require 10,000 vCPUs for 8 hours every night. The workload is fault-tolerant: if some instances are interrupted, the simulation can restart on new instances. The company uses Spot Fleet with the lowestPrice strategy to minimize costs. They specify multiple instance types (c5.4xlarge, c5a.4xlarge, m5.4xlarge) across three Availability Zones. Each night, the fleet launches from the cheapest pool. If that pool's price spikes, the fleet may fail to launch all capacity, but the company accepts this risk. They set maintain = false because the job is ephemeral. A common misconfiguration is using capacityOptimized here, which would increase costs unnecessarily.

A SaaS company runs a stateless web application on Spot Instances behind an Application Load Balancer. They need high availability and cannot tolerate all instances being interrupted at once. They use Spot Fleet with the diversified strategy across multiple instance types (t3.medium, t3a.medium, m5.large) and three AZs. The fleet distributes instances across the cheapest pools, ensuring that no single AZ or instance type carries all traffic. If one pool becomes unavailable, traffic shifts to the remaining pools. The company sets maintain = true to automatically replace interrupted instances. A common mistake is using lowestPrice, which would put all instances in one AZ, creating a single point of failure.

A media company needs to render 50,000 frames for a movie premiere in 2 hours. Cost is secondary; they need maximum compute capacity immediately. They use Spot Fleet with the capacityOptimized strategy, selecting multiple GPU instance types (p3.2xlarge, p3.8xlarge, g4dn.xlarge) across all AZs. The fleet picks the pool with the most available Spot capacity, launching all 50,000 instances as fast as possible. If a pool runs out, it moves to the next. The company sets maintain = true to ensure any interrupted frames are re-rendered. A misconfiguration would be using lowestPrice, which could delay the launch if the cheapest pool has limited capacity.