3.3 - High-performing compute (placement groups, instance types)
AWS Solutions Architect Associate objective 3.3 covers designing high-performing compute. A tightly-coupled HPC job that needs the lowest possible network latency and high throughput between instances benefits from a cluster placement group, which packs instances close together in one Availability Zone. Choosing the right instance family matters: a memory-heavy in-memory caching workload that spends little on CPU fits a memory-optimized family such as R, while compute-optimized C, general-purpose M, and accelerated (GPU) families suit other profiles. To get machine-learning-based recommendations on whether your EC2, EBS and Lambda resources are right-sized, AWS Compute Optimizer analyses utilisation and suggests changes. You should also know Auto Scaling and Graviton. Expect scenario questions that describe low-latency clustering, a resource profile, or right-sizing guidance, and ask which compute design or service - placement group, instance family or Compute Optimizer - fits.
Lowest latency between tightly-coupled instances = cluster placement group. Lots of RAM, little CPU = memory-optimized (R family). ML right-sizing recommendations for EC2/EBS/Lambda = Compute Optimizer.
Practice questions
1. A tightly-coupled HPC job needs the lowest possible network latency and highest packet-per-second rate between EC2 instances. Which placement strategy fits?
- A spread placement group distributing instances across distinct hardware
- A cluster placement group packing instances close within one Availability Zone (correct answer)
- A partition placement group isolating instances into separate racks
- No placement group, relying on enhanced networking alone across AZs
A cluster placement group packs instances onto close hardware in one AZ for low-latency, high-throughput node-to-node traffic, ideal for tightly-coupled HPC. Spread and partition prioritize fault isolation instead.
2. An in-memory caching workload spends heavily on RAM but little on CPU. Which EC2 instance family is the best fit?
- Memory-optimized instances such as the R family (correct answer)
- Compute-optimized instances such as the C family
- Storage-optimized instances such as the I family
- General-purpose burstable instances such as the T family
Memory-optimized R-family instances provide a high RAM-to-vCPU ratio, ideal for in-memory caches and databases. C is for compute-bound work, I for high local storage IOPS, T for low steady CPU.
3. A self-managed high-transaction database on EC2 needs sustained tens of thousands of IOPS with consistent single-digit-millisecond latency. Which EBS volume type is appropriate?
- io2 Block Express (Provisioned IOPS SSD) for high, consistent IOPS (correct answer)
- st1 (Throughput Optimized HDD) for its strong sequential throughput
- sc1 (Cold HDD) to keep the storage cost as low as possible
- gp2 (General Purpose SSD) relying on burst credits for peak IOPS
io2 Block Express delivers high, provisioned IOPS with consistent low latency and high durability, purpose-built for I/O-intensive, latency-sensitive databases. HDD types serve sequential, not transactional, workloads.
4. Nodes in a tightly-coupled HPC cluster need OS-bypass networking with very high packet rates and consistent low latency between instances. Which capability provides this?
- Attaching multiple standard elastic network interfaces to each node
- An Elastic Fabric Adapter (EFA) enabling OS-bypass HPC networking (correct answer)
- A Network Load Balancer distributing the inter-node cluster traffic
- A larger instance size so more network bandwidth is available per node
An Elastic Fabric Adapter provides OS-bypass networking (via libfabric) for very low, consistent latency and high packet rates between instances, purpose-built for tightly-coupled HPC and ML workloads.
5. A self-managed transactional database on EC2 needs consistent, very high provisioned IOPS with the highest durability among EBS options. Which volume type should be selected?
- gp3 General Purpose SSD sized up with extra provisioned throughput
- st1 Throughput Optimized HDD for its strong large-block sequential rate
- io2 Block Express Provisioned IOPS SSD for high IOPS and 99.999% durability (correct answer)
- sc1 Cold HDD to hold the transaction logs at the lowest storage cost
io2 Block Express delivers the highest provisioned IOPS with consistent sub-millisecond latency and 99.999% durability, purpose-built for critical, I/O-intensive transactional databases. gp3 tops out lower; HDDs suit sequential work.
6. A team wants machine-learning-based recommendations on whether their EC2, EBS, and Lambda resources are over- or under-provisioned. Which service provides this?
- AWS Trusted Advisor, which only lists broad best-practice checks by category
- Amazon CloudWatch, which surfaces the raw utilization metrics for each resource
- AWS Cost Explorer, which visualizes spending trends but not sizing guidance
- AWS Compute Optimizer, giving ML-based rightsizing recommendations per resource (correct answer)
Compute Optimizer analyzes historical utilization with machine learning to recommend optimal configurations for EC2 instances, Auto Scaling groups, EBS volumes, and Lambda functions, addressing both over- and under-provisioning.