Amazon EC2 Overview: Instances, Networking, Storage, and Workload Selection

Purpose of This Lesson

Amazon Elastic Compute Cloud (Amazon EC2) provides on-demand virtual servers in AWS data centers. This lesson focuses on the decisions an operator must make when deploying and managing EC2 instances: operating system and hardware profile, VPC and subnet placement, network interface and IP-address behavior, persistent versus ephemeral storage, lifecycle actions, and cost implications.

Key Concepts

• EC2 instance: A virtual server running in AWS. Instances can run Linux or Windows. macOS EC2 options use dedicated hardware rather than the normal virtual-server model.
• Instance family and size: Instance families are optimized for different workload characteristics, while the size determines the amount of allocated capacity. Examples include general purpose, compute optimized, memory optimized, accelerated computing, and storage optimized families.
• VPC and subnet placement: Every EC2 instance is launched into a subnet in a VPC. A subnet maps to one Availability Zone.
• Elastic Network Interface (ENI): The network adapter connecting an instance to the VPC. Each instance has at least one ENI and may have additional interfaces.
• Amazon EBS: Persistent, block-based storage attached to an instance over the network. EBS volumes must be in the same Availability Zone as the instance.
• Instance store: Physically attached host storage that offers high performance but is ephemeral. Data is lost when the underlying instance or host loses power.
• Instance lifecycle: Starting and stopping controls whether the instance is running; terminating deletes the instance. Running instances incur compute charges, while provisioned EBS capacity continues to incur storage charges whether the instance is running or stopped.

EC2 Architecture and Operational Responsibility

EC2 instances run as virtual machines on physical host servers managed by AWS. AWS manages the physical hosts and underlying data-center infrastructure. The customer manages the guest operating system, applications, configuration, patches, and security inside the virtual machine.

This is an important operational boundary: choosing EC2 does not transfer responsibility for the operating system or application stack to AWS. The customer must maintain and secure those components.

EC2 instance types provide different combinations of virtual CPUs, memory, storage connectivity, and network capacity. An instance family should be selected according to the workload rather than simply choosing the largest available size.

An identifier such as M5.large can be interpreted as follows:

• M: Instance family.
• 5: Generation.
• large: Instance size.

The lecture examples illustrate the capacity and price progression from smaller to larger instances: T2.micro, C3.2xlarge, and M4.16xlarge. Exact prices change over time, so these examples are useful for relative comparison rather than current pricing.

EC2 Networking in VPCs

EC2 instances are launched into subnets within a VPC. A VPC is a private networking boundary for an AWS environment. A subnet belongs to one Availability Zone, which represents a distinct AWS infrastructure location within a Region.

A subnet can be designed for private-only connectivity or public connectivity. Internet access requires the appropriate VPC networking configuration, including an internet gateway and a public IP address when direct public addressing is needed. Instances can also communicate with other networks, such as on-premises environments, when the surrounding network architecture provides that connectivity.

Each instance has at least one ENI. Additional ENIs can be attached, including ENIs in different subnets, provided those subnets are in the same Availability Zone as the instance. On Linux, the first interface is commonly visible as eth0; Windows uses different operating-system naming conventions.

Private and Public IP Addresses

An instance always has a private IP address associated with its ENI. In a public subnet, an instance may also receive a public IP address. A public IP address is dynamic and is lost when the instance is stopped and started again. A public IP is associated externally with the instance’s private address, so commands such as ifconfig or ipconfig show the private interface address rather than the public address association.

An Elastic IP address is a static public IPv4 address. It can be moved between instances or ENIs, making it appropriate when an endpoint must retain the same public address. Elastic IP addresses incur charges, as do public IP addresses according to the pricing conditions described in the lecture.

Do not confuse the behaviors:

• Dynamic public IP: Can change after a stop/start cycle.
• Private IP: Remains associated with the instance unless networking is reconfigured or the instance is replaced.
• Elastic IP: Static and movable between supported instances or ENIs.

Network Interface Types

• ENI: The standard network interface for normal networking requirements and available across EC2 instance types.
• ENA: Enhanced networking, providing higher bandwidth, higher performance, and lower inter-instance latency. It requires a supported instance type.
• EFA: Intended for high-performance computing, MPI, and machine-learning workloads that require very low latency for tightly coupled processing. Its use depends on supported EC2 configurations.

EC2 Storage Choices

Amazon EBS

Amazon EBS is persistent block storage. Although an EBS volume is attached over the network, the guest operating system presents it as a local block device. In Windows, it appears as a local drive rather than a network file share.

An EBS volume exists within a single Availability Zone and must be attached to an instance in that same Availability Zone. EBS volumes are automatically replicated within the Availability Zone for durability. EBS is the normal choice for data that must survive instance stop/start operations and for persistent boot or application storage.

EBS pricing is based on provisioned capacity. If a 100 GB volume is created, charges apply to the provisioned 100 GB even if only a portion contains data.

Important volume categories from the lesson include:

• gp3: General-purpose SSD for most workloads, including databases and development or test environments.
• gp2: An older general-purpose SSD option, commonly associated with boot volumes, development/test, and general use.
• io1 and io2: High-IOPS SSD options for workloads with substantial read/write demands. These are the EBS types identified in the lesson as supporting Multi-Attach.
• st1: Throughput-optimized HDD for workloads needing throughput rather than high IOPS.
• sc1: Cold HDD for lower-performance, lower-cost storage use cases.
• Magnetic: An older storage option that is not commonly used today.

Multi-Attach allows multiple instances to attach to one EBS volume, but it is not a general-purpose shared file system and has restrictions. The lecture identifies io1 and io2 as the volume types supporting this capability. For shared file storage, consider Amazon Elastic File System for Linux or Amazon FSx for Windows file servers instead.

Instance Store

Instance store volumes are physically attached to the EC2 host. They can provide very high performance, but they are ephemeral: data can be lost if power is lost or the instance is otherwise removed from the underlying host.

Use instance store only for data that can be recreated, temporary files, caches, or data replicated elsewhere. Persistent business data should be stored on EBS or another durable storage service. The performance benefit does not compensate for the risk of using instance store as the only copy of important data.

Exam- or Assessment-Relevant Takeaways

• Choose the instance family based on the workload ratio: compute optimized for relatively CPU-heavy work, memory optimized for relatively memory-heavy work, and other families for specialized requirements.
• Interpret instance names by family, generation, and size; a larger size generally provides more capacity and costs more.
• Remember that instances are launched into subnets, and subnets belong to one Availability Zone.
• Distinguish a dynamic public IP from an Elastic IP. A dynamic public IP can change after stop/start; an Elastic IP is static and movable.
• Every instance has at least one ENI. Multiple ENIs can be used, but interfaces must comply with Availability Zone placement constraints.
• EBS is persistent block storage and must be colocated with the instance in the same Availability Zone.
• EBS charges are based on provisioned capacity, not merely the amount of data currently stored.
• Instance store is high-performance but ephemeral; it is appropriate only when data can be recreated or is replicated elsewhere.
• ENA is for enhanced networking on supported instance types; EFA targets tightly coupled HPC, MPI, and machine-learning workloads.
• Multi-Attach is not equivalent to a shared file system. Use a file-system service when shared file semantics are required.
• Stopping an instance can reduce compute charges while leaving provisioned EBS storage charges in place. Terminating an instance deletes it.

Tool / Feature Decision Guide

Requirement	Appropriate choice	Decisive reason
Normal virtual server workload	EC2 instance selected from the appropriate family and size	Provides configurable CPU, memory, networking, and storage connectivity
CPU-heavy workload	Compute-optimized family	Higher CPU-to-memory balance
Memory-heavy workload	Memory-optimized family	Higher memory-to-CPU balance
Persistent block storage	Amazon EBS	Data persists independently of normal instance running state
Temporary, recreatable, high-performance data	Instance store	Very high performance, with an accepted loss-of-data risk
General-purpose SSD storage	gp3	Suitable for most workloads, including databases and development/test
Older general-purpose SSD deployment	gp2	Legacy general-purpose option described in the lesson
High IOPS requirement	io1 or io2	Designed for heavier read/write workloads
High throughput at lower cost	st1	Throughput-optimized HDD rather than high-IOPS storage
Low-cost, lower-performance HDD storage	sc1	Cold HDD option
Stable public address	Elastic IP	Static address that can move between instances or ENIs
Enhanced network performance	ENA on a supported instance	Higher bandwidth and lower inter-instance latency
Tightly coupled HPC, MPI, or ML workload	EFA on a supported configuration	Designed for very low-latency communication
Shared file storage	EFS for Linux or FSx for Windows	Multi-Attach EBS is block storage, not a shared file system

Common Traps / Misconceptions

• Stopping is not terminating. Stopping preserves the instance for later restart; terminating deletes it.
• Stopping an instance does not eliminate all charges. Running-instance charges stop, but provisioned EBS capacity remains billable.
• An unused EBS allocation is still charged. Billing is based on provisioned capacity, not the amount of data written.
• A public subnet does not automatically mean a stable public endpoint. A dynamic public IP can change after a stop/start cycle; use an Elastic IP when stability is required.
• The public IP is not shown as a normal interface address inside the guest. The operating system primarily sees the private address on the ENI.
• EBS is not a network file share. It is block storage attached to an instance.
• Multi-Attach is not general shared storage. It has restrictions and should not be treated as a replacement for EFS or FSx.
• Instance store is not persistent just because it looks like a local disk. Its physical attachment provides performance, not durability.
• The largest instance is not automatically the best instance. Overprovisioning increases cost; match the family and size to workload needs.
• ENA and EFA are not interchangeable. ENA is enhanced networking, while EFA is aimed at specialized low-latency HPC, MPI, and machine-learning communication.

Real-World Engineer / Analyst Notes

• Begin EC2 sizing with observed workload characteristics: CPU utilization, memory pressure, I/O behavior, and network demand. Revisit the size as the workload changes.
• Design public exposure deliberately. Prefer private subnets when direct internet addressing is not required, and use the surrounding VPC architecture to control connectivity.
• Treat an Elastic IP as a dependency that must be documented. Moving a static address can preserve endpoint identity, but it does not by itself solve application failover or availability design.
• Separate persistent data from disposable data. Put operating-system, application, and business data that must survive on EBS or another durable service; reserve instance store for caches, scratch data, or replicated content.
• Check Availability Zone placement before attaching EBS volumes or additional ENIs. A mismatch prevents the intended attachment.
• For automated operations, EC2 lifecycle actions can be performed through the AWS console, command-line tools, SDKs, or automation processes. Automation should distinguish a recoverable stop from irreversible termination.
• Review current AWS documentation and pricing when implementing a design. Instance families, volume options, capabilities, and prices can change over time.

Quick Reference Summary

• EC2 provides on-demand virtual servers; Linux and Windows use the virtual-server model, while macOS uses dedicated hardware.
• AWS manages physical hosts; the customer manages the guest operating system, applications, patches, and security.
• Instances run in VPC subnets, and each subnet belongs to one Availability Zone.
• Every instance has at least one ENI; additional ENIs are possible within placement constraints.
• Dynamic public IPs can change after stop/start. Elastic IPs are static and movable.
• EBS is persistent, block-based, same-AZ storage charged by provisioned capacity.
• Instance store is physically attached, high-performance, and ephemeral.
• Select instance families and sizes according to CPU, memory, storage, and networking requirements.
• ENA provides enhanced networking; EFA targets specialized low-latency HPC, MPI, and ML workloads.
• Multi-Attach is limited block storage functionality, not a shared file system.

Flashcards

Q: A workload is CPU-heavy and does not require an unusually large memory footprint. Which EC2 family direction should you investigate, and why?
A: Investigate a compute-optimized family because it provides a higher CPU-to-memory balance than a memory-optimized family.

Q: What do the three parts of M5.large represent?
A: M is the instance family, 5 is the generation, and large is the instance size.

Q: An instance must retain the same public IP address after being stopped and started. Which feature should you choose?
A: Use an Elastic IP address because it is static and can be associated with the instance again after a lifecycle change.

Q: What is the key difference between a dynamic public IP and an Elastic IP?
A: A dynamic public IP can be lost after a stop/start cycle, while an Elastic IP is static and can be moved between supported instances or ENIs.

Q: An instance needs two network interfaces in different subnets. What placement constraint applies?
A: The subnets must be in the same Availability Zone as the instance.

Q: Why might an EC2 guest operating system not display its public IP when you inspect the network interface locally?
A: The public IP is associated externally with the instance’s private IP; the guest operating system sees the ENI and its private address.

Q: When should EBS be selected instead of instance store?
A: Select EBS when data must be persistent. Instance store should be limited to temporary, recreatable, or replicated data because it is ephemeral.

Q: An application needs high read/write performance and very high IOPS. Which EBS category from the lesson is the best starting point?
A: Consider io1 or io2 because they are intended for high-IOPS workloads.

Q: How does EBS billing treat a provisioned but mostly unused volume?
A: Charges are based on the provisioned capacity, so a 100 GB volume is charged as 100 GB even if little data is stored.

Q: Why is Multi-Attach not a substitute for EFS or FSx?
A: Multi-Attach still provides block storage with restrictions; EFS or FSx is the more appropriate direction when shared file-system semantics are needed.

Q: Which networking technology is associated with tightly coupled HPC, MPI, and machine-learning workloads?
A: Elastic Fabric Adapter (EFA), which is designed for very low-latency communication in those specialized workloads.

Q: What is the cost-related difference between stopping and terminating an instance?
A: Stopping ends running-instance charges but leaves the instance recoverable and does not remove EBS charges. Terminating deletes the instance and is not the normal reversible shutdown action.

Practice Questions

Question 1

A development server is used only during business hours. The team wants to reduce compute costs overnight, but its EBS boot and application volumes must remain available for the next morning. What should the operator do?

A. Terminate the instance every evening and recreate it manually
B. Stop the instance overnight and leave the EBS volumes provisioned
C. Delete the EBS volumes while leaving the instance running
D. Move the instance store data to a public IP address

Correct answer: B

Stopping avoids charges for the running instance while preserving the instance and its EBS-backed data. Provisioned EBS capacity continues to incur storage charges.

Question 2

A service must use the same public IPv4 address after planned EC2 stop/start operations. Which option best meets the requirement?

A. A dynamic public IP assigned at launch
B. The instance’s private IP address only
C. An Elastic IP address
D. An additional instance-store volume

Correct answer: C

An Elastic IP is static and can remain associated with the service across stop/start operations. A dynamic public IP may change.

Question 3

A financial application needs persistent block storage attached to an EC2 instance. The required EBS volume currently exists in a different Availability Zone from the instance. What is the decisive issue?

A. EBS volumes must be in the same Availability Zone as the instance
B. EBS can attach across any Availability Zone without configuration
C. Only instance store can be attached across Availability Zones
D. The volume must first be converted to a public subnet

Correct answer: A

The instance and EBS volume must be in the same Availability Zone for attachment. Subnet public/private status does not solve an Availability Zone mismatch.

Question 4

A tightly coupled machine-learning workload exchanges data frequently between compute nodes and requires very low inter-instance latency. Which networking option is the best fit from the lesson?

A. A standard ENI only
B. An EFA on a supported configuration
C. A dynamic public IP address
D. A cold HDD volume

Correct answer: B

EFA is intended for high-performance computing, MPI, and machine-learning use cases with tightly coupled, low-latency communication.

Question 5

An operations team wants several EC2 instances to access the same data and assumes EBS Multi-Attach is equivalent to a shared file system. What is the correct guidance?

A. Use any EBS type because all support Multi-Attach
B. Use instance store because it is automatically shared by host instances
C. Use io1 or io2 only for the described Multi-Attach capability, and evaluate EFS or FSx for shared files
D. Assign each instance a different Elastic IP to create shared storage

Correct answer: C

The lesson identifies io1 and io2 as supporting Multi-Attach, but Multi-Attach remains block storage with restrictions. EFS for Linux or FSx for Windows is more appropriate for shared file-system requirements.

WordPress Metadata

Suggested Slug:
amazon-ec2-overview-instances-networking-storage

Meta Description:
Study Amazon EC2 instance types, VPC networking, IP addresses, EBS and instance store, lifecycle states, and workload-selection tradeoffs for AWS CloudOps Engineer Associate preparation.

Tags:
Amazon EC2, EC2 instance types, Amazon VPC, Elastic Network Interface, Elastic IP address, Amazon EBS, Instance store, AWS storage, AWS networking, AWS CloudOps