KodeKloud Notes

AWS EC2 Placement Groups let you control instance placement on the underlying hardware to meet performance and availability goals. By grouping instances, you can achieve low-latency, high-throughput networking or isolate workloads against rack-level failures.

Suppose you deploy a Spark cluster in the Singapore region across multiple Availability Zones. Inter–AZ latency can reach up to 10 ms—too high for tightly coupled HPC workloads. Placement groups solve this by selecting optimal racks for your EC2 instances.

The image shows a diagram illustrating the need for placement groups, with three locations labeled ap-southeast-1A, ap-southeast-1B, and ap-southeast-1C, each connected by lines indicating a 10 ms latency.

What Is an AWS Placement Group?

An AWS Placement Group is a logical construct that influences the physical placement of EC2 instances to optimize for:

Low network latency and high throughput
Hardware-failure isolation
Minimized blast radius for critical services

Choose from three placement group strategies:

Strategy	Description	Best for
Cluster	Packs instances closely in one AZ for low latency & high throughput	HPC, tightly coupled compute
Partition	Divides instances into partitions with separate racks	Large distributed systems requiring fault isolation
Spread	Distributes a small number of instances across racks or AZs	Critical components that need failure containment

The image illustrates different types of placement groups: Cluster, Partition, and Spread, each within an availability zone, showing how instances are organized.

Below is an in-depth look at each placement group type.

Cluster Placement Groups

A cluster placement group co-locates instances within a single Availability Zone. This yields ultra-low latency and high network throughput, ideal for HPC clusters or big-data workloads.

Key requirements and benefits:

Instances must support enhanced networking (ENA or Elastic Fabric Adapter)
All instances reside in the same AZ (no cross-AZ placement)
Network throughput can exceed 10 Gbps on supported instance types
Use the same instance type within the group for maximum performance

Note

To leverage the highest network performance, ensure your AMI and instance types support the Elastic Network Adapter (ENA) or Elastic Fabric Adapter (EFA).

The image illustrates a "Placement Group - Cluster" concept, showing enhanced networking within an availability zone, with a focus on network speeds and connectivity between different nodes.

Partition Placement Groups

A partition placement group segments your fleet into logical partitions. Each partition is mapped to its own set of racks with isolated power and network infrastructure, reducing the risk of simultaneous hardware failures.

Highlights:

Can span multiple Availability Zones in a single region
Up to 7 partitions per AZ, each on separate racks
Racks in different partitions do not share power or network gear

Warning

Do not exceed 7 partitions per Availability Zone. If you need more partitions, consider multiple placement groups.

Spread Placement Groups

A spread placement group distributes a small number of instances across distinct racks, AZs, or both. This strategy minimizes the blast radius of hardware failures for critical application components.

Key limits:

Can span multiple AZs within a region
Maximum 7 running instances per AZ per spread group
In a region with 3 AZs, you can have up to 21 instances in one spread group

The image illustrates a "Spread Placement Group" in a cloud computing context, showing two availability zones with distributed instances. It highlights the concept of spreading instances across different racks with their own network and power sources.

Links and References

Watch Video

Watch video content