Read on:
AWS for Beginners: Understanding AWS Terminologies: Part 1
AWS for Beginners: What is AWS EC2: Part 2
AWS for Beginners: AWS Compute Services: Part 3
AWS for Beginners: How to Protect AWS Security Tools: Part 4
AWS for Beginners: What is IAM (Identity and Access Management) and Best Practices: Part 5
AWS for Beginners: Provisioning IAM Users and Programmatic Access in AWS: Part 6
AWS for Beginners: Securing Root account using MFA: Part 7
AWS for Beginners: Amazon S3 Overview, Security and Best Practices : Part 8
AWS for Beginners: Creating and managing IAM groups: Part 9
AWS for Beginners: AWS Managed Policies and In-line Policies: Part 10
AWS for Beginners: Create and Attach a policy to an IAM User: Part 11
AWS for Beginners: How to Create a Customer Managed Policy: Part 12
Placement group strategy
Amazon Web Services (AWS) provides various services to set up high-performance computing (HPC) environments on the go. EC2 instance types like CPU optimized, GPU optimized, Spot instances/ Spot fleets and auto-scaling helps to form high-performance computing at optimal cost. By controlling the EC2 instances placement, you increase the performance or improve the availability. This article will walk you through the different types of AWS EC2 placement groups and how they can be set up in the AWS cloud environment.
Placement Group Cluster:
If you are looking for ultra-low network (10Gbps network) latency between instances in a distributed fashion, then a placement group cluster is a great option. The following diagram shows how the cluster placement group will be set up in AWS. Placement group cluster instances will be on the same rack and same availability zone. EC2 instances can benefit from faster network speed but it comes with a high risk. Failure of a single rack will result in the loss of all EC2 instances. It suits applications that require fast processing and are not critical.
Pros:
1. Ultra-low latency between EC2 instances since it’s placed on the same Availability Zone and same rack.
2. It offers the highest network throughput up to 10Gbps
3. It’s an ideal option for fast processing and faster communication between instances.
Cons:
1. Single rack failure can bring down the whole environment
2. High risk since instances are placed on the same rack and same AZ.
Spread Placement Group:
The Spread Placement group allows EC2 instances over multiple availability zones. This strategy will ultimately put the instances into different hardware which significantly reduces the risk. The spread placement group has limitations where you can have only 7 instances per Availability Zone per placement group. This is an ideal placement strategy for critical applications where failure is not an option.
Pros:
1. Provides High availability.
2. Reduces the risk of failure and ideal for critical applications.
Cons:
1. Only 7 instances can be part of the placement group from one AZ.
2. Compared to a cluster placement group, network performance will be lesser since instances are spread across multiple AZs.
Partition Placement Group:
The Partition placement group overcomes the high availability issues of the cluster placement group and the number of instances limitation issues of the spread placement group. This new type of placement group strategy allows 100s of instances to be placed in a single AZ but on a different rack. Server racks are divided into logical partitions and we can have 7 partitions per AZ. The following diagram illustrates the partition placement group strategy.
Pros:
1. Provides High availability.
2. Good for distributed and replicated workloads. (Example: HDFS, HBase, Cassandra)
Cons:
1. Medium risk associated with it. It has low risk compared to cluster placement group but higher risk compared to spread placement group.
Conclusion:
AWS EC2 placement group strategies should be considered while designing the low latency applications and critical workloads. It provides customers to choose where they want to place instances based on the criticality of the workloads. Cluster, Spread, and partition placement have their own pros and cons. The architect needs to access the workload before choosing the placement group type. Kindly note that you can’t merge placement groups with each other. An instance can be part of one placement group at a time.
Follow our Twitter and Facebook feeds for new releases, updates, insightful posts and more.
