5.1 Compute Services Deep Dive
Key Takeaways
- Elastic Beanstalk handles capacity provisioning, load balancing, auto-scaling, and health monitoring — you just upload code.
- AWS Batch dynamically provisions EC2 or Spot instances to run batch computing jobs without manual cluster management.
- Amazon Lightsail is the simplest way to launch a virtual private server — ideal for simple websites and small applications.
- EC2 Instance Store provides temporary block-level storage physically attached to the host — data is lost when the instance stops.
- Placement Groups control EC2 instance placement: Cluster (low latency), Spread (high availability), and Partition (large distributed systems).
Quick Answer: This deep dive sharpens compute distinctions the exam loves to probe: Elastic Beanstalk vs. Lightsail vs. EC2, instance store (ephemeral) vs. EBS (persistent), Dedicated Hosts vs. Dedicated Instances, and the advanced purchasing and load-balancing options layered on top of EC2.
AWS Elastic Beanstalk — PaaS for Web Applications
Elastic Beanstalk is the easiest way to deploy and run web applications on AWS. You upload code and Beanstalk handles capacity provisioning, load balancing, auto-scaling, health monitoring, and platform patching automatically.
Supported platforms: Java, .NET, PHP, Node.js, Python, Ruby, Go, and Docker.
Exam-relevant facts:
- Free service — you pay only for the underlying resources (EC2, ELB, RDS)
- You keep full control of the resources it creates
- Ideal for developers who want to focus on code, not infrastructure
- Supports multiple environments (production, staging, development)
On the Exam: Elastic Beanstalk = "deploy my web app without managing infrastructure." It is the canonical PaaS answer on AWS.
AWS Batch
AWS Batch plans, schedules, and runs batch computing jobs at any scale. You define jobs, job definitions, and compute environments; Batch then provisions the optimal quantity and type of EC2 or Spot instances, runs the work in Docker containers, and scales the fleet down when finished. Typical uses include genomics, financial risk modeling, media transcoding, and large simulations.
Amazon Lightsail
Amazon Lightsail is the simplest on-ramp to AWS: bundled virtual servers, storage, databases, and networking at a predictable monthly price.
| Feature | Detail |
|---|---|
| Pricing | Fixed monthly, starting at a low flat rate |
| Use Cases | Simple websites, blogs, small apps |
| Includes | Compute, storage, and transfer in one bundle |
| Management | Simplified compared to raw EC2 |
On the Exam: "Simple WordPress blog with predictable pricing and minimal management" = Lightsail, not full EC2 with Auto Scaling.
EC2 Storage Options
Instance Store vs. EBS
| Feature | Instance Store | EBS |
|---|---|---|
| Persistence | Ephemeral — data lost on stop/terminate | Persistent — survives stop/start |
| Performance | Very high (physically attached) | High (network-attached SSD/HDD) |
| Backup | Cannot snapshot | Snapshots stored in S3 |
| Use Case | Caches, buffers, scratch data | Boot volumes, databases, persistent data |
On the Exam: Instance Store = ephemeral. EBS = persistent. If a scenario requires data to survive instance termination, the answer involves EBS or S3 — never instance store.
EC2 Purchasing Options (Advanced)
Spot Fleet
A Spot Fleet launches a target capacity from a mix of Spot (and optionally On-Demand) instances, automatically drawing from the lowest-priced capacity pools to meet your goal at minimal cost.
On-Demand Capacity Reservations
Capacity Reservations guarantee EC2 capacity in a specific AZ when you need it. Unlike Reserved Instances there is no term commitment, but you pay On-Demand rates whether or not the capacity is used — useful for guaranteed capacity during predictable peak events.
Dedicated Hosts vs. Dedicated Instances
| Feature | Dedicated Host | Dedicated Instance |
|---|---|---|
| Billing | Per host | Per instance |
| Server Visibility | You see the physical server (sockets/cores) | You do not |
| BYOL Support | Yes (socket/core-bound licenses) | No |
| Placement Control | Yes | No |
| Use Case | BYOL licensing, strict compliance | Hardware isolation, compliance |
On the Exam: "Reuse existing per-socket/per-core licenses" = Dedicated Hosts (they expose the physical hardware). Dedicated Instances isolate hardware but do not expose socket/core detail.
Elastic Load Balancing Deep Dive
Health Checks
Every ELB type performs health checks on registered targets and automatically removes unhealthy ones from rotation until they recover — a core building block of high availability.
Cross-Zone Load Balancing
Distributes traffic evenly across all targets in all enabled AZs, even when AZs hold unequal numbers of instances, preventing hotspots.
Connection Draining (Deregistration Delay)
When a target is deregistering or unhealthy, the load balancer lets in-flight requests finish for a configurable period (default 300 seconds) before fully removing it, avoiding dropped user sessions during scale-in or deployments.
On the Exam: Pair these with Auto Scaling: ELB health checks plus Auto Scaling replacement plus Multi-AZ deployment is the standard recipe behind any "highly available" web-tier question.
Choosing Among Compute Options
The deep-dive services map to a simple spectrum of control vs. convenience. At the most-control end sits EC2, where you own the OS and everything on it. Elastic Beanstalk sits one step toward convenience: still EC2 underneath, but AWS automates provisioning, scaling, and health. Lightsail trades flexibility for a fixed monthly price and the simplest experience. AWS Batch is purpose-built for queued, large-scale jobs rather than always-on services, and at the far convenience end sit Lambda and Fargate, where there are no servers to manage at all.
A reliable exam approach is to read for the operational burden the customer wants to avoid. "We just want predictable pricing and minimal setup" points to Lightsail. "We want to deploy a web app and let AWS handle scaling" points to Elastic Beanstalk. "We need to run thousands of independent batch jobs" points to AWS Batch. "We do not want to manage any servers" points to Lambda or Fargate. Matching the amount of management the customer wants to give up to the right service resolves most compute questions quickly.
Placement Groups
EC2 placement groups influence how instances are placed on underlying hardware. A Cluster placement group packs instances close together for the lowest network latency (HPC, tightly coupled compute). A Spread placement group puts each instance on distinct hardware to maximize availability for a small number of critical instances. A Partition placement group spreads instances across logical partitions so large distributed systems (such as Hadoop or Cassandra) limit the blast radius of a hardware failure. The exam may ask which placement group fits a latency-sensitive cluster versus a failure-isolated distributed system.
Which AWS service automatically handles capacity provisioning, load balancing, and auto-scaling for web applications, so developers can focus solely on their code?
What happens to data stored on an EC2 Instance Store volume when the instance is stopped?
A small business wants to host a simple WordPress blog on AWS with predictable monthly pricing. Which service is most appropriate?
Which EC2 option should a company choose if they need to use their existing Windows Server licenses that are tied to physical server cores?