This chapter covers EC2 instance type families and their use cases, a core topic for the SAA-C03 exam. Understanding which instance family to choose for a given workload is critical for designing cost-effective and performant architectures. Approximately 10-15% of exam questions directly test instance selection, often in the context of workload requirements like compute, memory, or storage optimization. Mastering this topic ensures you can architect resilient, scalable solutions that align with AWS best practices.
Jump to a section
Imagine a delivery company that owns a fleet of vehicles. Each vehicle type is designed for specific tasks: a sports car (Compute Optimized) has a powerful engine but minimal cargo space—perfect for fast, single-item deliveries. A pickup truck (General Purpose) balances engine power and cargo capacity, handling most daily routes. A heavy truck (Memory Optimized) has a huge cargo hold for bulky items but is slower. A refrigerated truck (Storage Optimized) has massive storage with temperature control for perishables. A hybrid car (Accelerated Computing) has specialized electric motors for high-torque tasks like towing. Each vehicle type has a different cost per mile, fuel efficiency, and maintenance schedule. When a package arrives, the dispatcher selects the vehicle that best matches the package's size, urgency, and special handling needs. If they use a sports car for a refrigerator, it fails. If they use a heavy truck for a letter, they waste money. Similarly, EC2 instance families are specialized virtual machines; choosing the wrong family leads to poor performance or overspending. The exam tests your ability to match workload characteristics to the correct instance family, just as a dispatcher matches cargo to vehicle.
What Are EC2 Instance Families?
Amazon EC2 provides a broad selection of instance types, each optimized for different use cases. Instance types are grouped into families based on their resource ratios—CPU, memory, storage, and networking. The families are: General Purpose, Compute Optimized, Memory Optimized, Accelerated Computing, Storage Optimized, and HPC Optimized. Each family contains multiple instance sizes (e.g., t3.micro, c5.xlarge) that scale resources linearly. The naming convention is: [family][generation].[size], where family is a letter code (e.g., t for General Purpose, c for Compute Optimized, r for Memory Optimized, i for Storage Optimized, p for Accelerated Computing) and generation is a number (e.g., 5, 6). Sizes range from nano to 24xlarge, with some families offering metal sizes.
Why Do Instance Families Exist?
No single instance type can efficiently handle all workloads. Some applications require high CPU clock speeds for batch processing, while others need massive memory for in-memory databases. By offering families with different ratios, AWS allows you to match your workload's dominant resource requirement. This prevents overprovisioning (paying for unused resources) and underprovisioning (performance bottlenecks). The exam expects you to identify the correct family for a given scenario, such as choosing a Compute Optimized instance for video encoding or a Memory Optimized instance for large caches.
How Instance Families Work Internally
Each instance family is built on a specific combination of underlying hardware, including CPU type (Intel Xeon, AMD EPYC, or AWS Graviton), memory technology (DDR4 vs DDR5), and storage (EBS vs instance store). For example, the C5 family uses Intel Xeon Platinum processors with up to 3.6 GHz sustained Turbo frequency, while the C6g family uses AWS Graviton2 ARM-based processors. The allocation of resources is managed by the Nitro hypervisor, which provides near-bare-metal performance by offloading virtualization functions to dedicated hardware. This means each instance type has a fixed ratio of vCPUs to memory. For instance, a t3.medium has 2 vCPUs and 4 GiB memory, giving a 1:2 ratio. Understanding these ratios helps you predict behavior: a Compute Optimized instance like c5.xlarge has 4 vCPUs and 8 GiB memory (1:2 ratio) but with higher CPU performance, while a Memory Optimized instance like r5.xlarge has 4 vCPUs and 32 GiB memory (1:8 ratio).
Key Components and Defaults
vCPU: Each vCPU is a thread of a physical CPU core. For most instances, 1 vCPU = 1 hyperthread. Exceptions: instances with "large" in the name may have 2 vCPUs per core.
Memory: Measured in GiB. The ratio of memory to vCPU is a key differentiator.
EBS Bandwidth: Varies by instance size. For example, a t3.nano has up to 0.208 Gbps, while a t3.2xlarge has up to 4.5 Gbps.
Network Bandwidth: Also varies by size. Smaller instances have burstable network performance via the AWS Nitro system.
Instance Store: Some families (e.g., i3, d2) include NVMe SSD instance store volumes for temporary, high-speed storage. Data is lost on stop/termination.
Burstable Performance: T family instances (e.g., t3, t4g) accumulate CPU credits when idle and consume them when active. This allows cost-effective baseline performance with bursts. The default credit balance is 144 credits for t3.nano, and the earn rate is 6 credits per hour per vCPU.
Configuration and Verification
To list available instance types, use the AWS CLI:
aws ec2 describe-instance-types --filters "Name=instance-type,Values=t3.*"To view running instances and their types:
aws ec2 describe-instances --query 'Reservations[*].Instances[*].[InstanceId,InstanceType]'When launching an instance, you specify the type in the console, CLI, or SDK. For example:
aws ec2 run-instances --image-id ami-0abcdef1234567890 --instance-type t3.micro --key-name MyKeyPairYou can also change the instance type after launch (stop required for most types) using:
aws ec2 modify-instance-attribute --instance-id i-1234567890abcdef0 --instance-type "{\"Value\": \"t3.small\"}"Interaction with Related Technologies
Amazon EBS: Instance types have different maximum IOPS and throughput for EBS volumes. For example, a t3.nano can achieve up to 100 MB/s for EBS, while a t3.2xlarge can achieve up to 2040 MB/s. Provisioned IOPS volumes (io1/io2) require instances with EBS-optimized capability.
VPC Networking: Enhanced networking (SR-IOV) is supported on most current instance types, providing higher packet-per-second performance. The Elastic Network Adapter (ENA) is used for speeds up to 100 Gbps.
Auto Scaling: When configuring Auto Scaling groups, you can specify multiple instance types to diversify risk and optimize cost. This is often tested in the exam for fault tolerance.
Spot Instances: Any instance family can be used as a Spot Instance, but certain families (like p3 for GPU workloads) are more commonly used due to cost savings. The exam may ask about choosing a family for a Spot-based workload.
General Purpose (A1, T3, T4g, M5, M6g, M6i, M7g)
These instances offer a balanced ratio of CPU, memory, and networking. They are ideal for workloads with moderate usage patterns, such as web servers, small databases, and development environments. The T family (burstable) is cost-effective for workloads that do not require sustained high CPU. The M family (standard) provides consistent performance. For example, a t3.micro (1 vCPU, 1 GiB memory) costs ~$0.0104 per hour, while an m5.large (2 vCPU, 8 GiB memory) costs ~$0.096 per hour. Use T instances for microservices with variable load, and M instances for production web servers.
Compute Optimized (C5, C5a, C5n, C6g, C6gn, C7g)
These instances have a high ratio of vCPUs to memory (e.g., 1:2) and are designed for compute-intensive workloads. Use cases include batch processing, scientific modeling, gaming servers, and video encoding. The C5 family offers up to 3.6 GHz sustained clock speed. For example, a c5.xlarge (4 vCPU, 8 GiB memory) costs ~$0.17 per hour. The C6g family uses Graviton2 processors, offering up to 40% better price-performance for some workloads. Choose Compute Optimized when the workload is CPU-bound and memory requirements are modest.
Memory Optimized (R5, R5a, R5b, R5n, R6g, R6i, R7g, X1, X1e, High Memory)
These instances have a high memory-to-vCPU ratio (e.g., 1:8 or higher). They are designed for memory-intensive applications such as in-memory databases (Redis, Memcached), real-time analytics, and large caches. For example, an r5.xlarge (4 vCPU, 32 GiB memory) costs ~$0.252 per hour. The X1e family offers up to 3,904 GiB of memory (x1e.32xlarge). High Memory instances (u-6tb1, u-9tb1, u-12tb1) provide up to 12 TB of memory for SAP HANA. Choose Memory Optimized when the workload requires large datasets to reside in RAM.
Accelerated Computing (P3, P4, Inf1, Trn1, G3, G4ad, G4dn, G5, G5g, VT1)
These instances include hardware accelerators (GPUs, FPGAs, AWS Inferentia, AWS Trainium) for tasks like machine learning training and inference, graphics rendering, and video transcoding. For example, p3.2xlarge (1 NVIDIA Tesla V100 GPU, 8 vCPU, 61 GiB memory) costs ~$3.06 per hour. The Inf1 family uses AWS Inferentia chips for low-cost, high-throughput inference. Choose Accelerated Computing when the workload benefits from parallel processing or specialized hardware.
Storage Optimized (I3, I3en, D2, D3, D3en, H1)
These instances provide high, sequential I/O for local storage. Use cases include NoSQL databases (Cassandra, MongoDB), data warehousing, and log processing. The I3 family offers NVMe SSD instance store with up to 15.2 TB of storage. D3 instances provide HDD storage for large, sequential workloads. For example, i3.large (2 vCPU, 15.25 GiB memory, 1 x 475 GB NVMe SSD) costs ~$0.156 per hour. Choose Storage Optimized when the workload demands high disk throughput and low latency, and data can be ephemeral.
HPC Optimized (Hpc6a, Hpc7g)
These instances are designed for tightly coupled, high-performance computing (HPC) workloads that require high network bandwidth and low latency. They use Elastic Fabric Adapter (EFA) for inter-node communication. For example, hpc6a.48xlarge (96 vCPU, 384 GiB memory) provides 100 Gbps network bandwidth. Choose HPC Optimized for scientific simulations, computational fluid dynamics, and weather modeling.
Instance Size and Scaling
Within a family, larger instances offer linearly more resources. For example, a c5.large has 2 vCPU and 4 GiB memory, while a c5.4xlarge has 16 vCPU and 32 GiB memory. Larger sizes also have higher network and EBS bandwidth. The exam may test that certain features (like EBS optimization) are only available on certain sizes. For instance, t3.nano, t3.micro, and t3.small are not EBS-optimized by default, but they can be enabled at no extra cost (subject to baseline performance).
Choosing the Right Family: A Decision Framework
Identify the bottleneck resource: CPU, memory, storage, or network.
If CPU-bound and memory is secondary → Compute Optimized.
If memory-bound (large datasets in RAM) → Memory Optimized.
If storage I/O is the primary requirement → Storage Optimized.
If workload benefits from GPUs or specialized hardware → Accelerated Computing.
If workload is balanced or variable → General Purpose (T for burstable, M for steady).
If workload is tightly coupled HPC → HPC Optimized.
Exam Example
Question: A company runs a real-time analytics dashboard that processes streaming data in memory. The workload requires high memory bandwidth and large amounts of RAM. Which instance family is most suitable? Answer: Memory Optimized (e.g., R5). Common wrong answer: Compute Optimized (because analytics sounds compute-heavy). The key is "in memory" indicating memory is the bottleneck.
Identify Workload Bottleneck
Begin by analyzing the workload's primary resource constraint: CPU, memory, storage I/O, or network throughput. For example, a batch processing job that consumes 100% CPU indicates a CPU bottleneck. A database that caches all data in memory indicates a memory bottleneck. This step determines which instance family to focus on. If the workload is balanced, General Purpose is a safe default. The exam often provides clues like "CPU-intensive" or "large in-memory cache" to guide this decision.
Select Instance Family
Based on the bottleneck, choose the appropriate family: Compute Optimized (C series) for CPU, Memory Optimized (R, X series) for memory, Storage Optimized (I, D series) for high disk I/O, Accelerated Computing (P, G, Inf series) for GPU/FPGA workloads, or General Purpose (T, M series) for balanced or variable loads. For example, a video transcoding service is CPU-bound, so select Compute Optimized. The exam may test that T series is not suitable for sustained CPU loads due to credit exhaustion.
Determine Instance Size
After choosing the family, select a size that meets the required vCPU and memory. Use the instance type naming convention: family.generation.size. For example, c5.xlarge provides 4 vCPU and 8 GiB memory. If the workload requires 16 vCPU, choose c5.4xlarge. Consider that larger sizes have higher network and EBS bandwidth. For burstable instances, ensure the credit balance can sustain the load. For instance, t3.medium has 2 vCPU and 4 GiB memory but only 0.208 Gbps baseline network; burst bandwidth is up to 5 Gbps.
Configure Storage and Networking
Attach appropriate EBS volumes (gp3, io1, etc.) or use instance store if available. For high I/O, choose Storage Optimized instances with NVMe instance store. Enable EBS optimization for consistent performance. For networking, enable enhanced networking (ENA) and assign an Elastic IP if needed. For HPC, use EFA. For example, a database on i3.large uses the local NVMe SSD for high-throughput logs. The exam may ask about instance store persistence: data is lost on stop/terminate.
Launch and Verify Performance
Launch the instance using the chosen type. Monitor performance metrics via CloudWatch (CPUUtilization, MemoryUtilization, DiskReadOps). If performance is insufficient, resize to a larger instance within the same family or switch families. For example, if a c5.xlarge shows high memory usage, consider an r5.xlarge. The exam may test that you can change instance type only when the instance is stopped (for most types) and that the new type must be compatible with the current network and storage configuration.
Enterprise Scenario 1: E-Commerce Web Application
A large e-commerce platform runs a web application on a fleet of EC2 instances. The workload is variable: during normal hours, CPU usage is around 20%, but during flash sales, it spikes to 80%. The company uses T3 instances (t3.large) for the web tier because of burstable performance. They configure Auto Scaling to add instances during sales. However, they discovered that sustained high CPU depletes CPU credits, causing performance degradation. To fix this, they switched to M5 instances for the web tier during sales, using a mixed instances policy in Auto Scaling. They also use R5 instances for the session cache (ElastiCache Redis) because it requires large memory. This architecture saves 30% cost compared to using M5 instances all the time. Common misconfiguration: using T instances for a database server, which leads to credit exhaustion and latency spikes. The exam tests that T instances are not suitable for workloads with sustained high CPU.
Enterprise Scenario 2: Machine Learning Training
A research lab trains deep learning models on large datasets. They use P3 instances (p3.16xlarge) with 8 NVIDIA V100 GPUs for training. The training job runs for days, and they use Spot Instances to reduce cost by 70%. They also use FSx for Lustre as a high-performance file system to store training data. A common mistake is using Compute Optimized instances (C5) for GPU training, which lacks the necessary parallel processing power. The exam may ask which instance family is best for ML training: Accelerated Computing (P3, P4, Trn1).
Enterprise Scenario 3: High-Performance Computing (HPC)
A weather forecasting organization runs computational fluid dynamics simulations. They use Hpc6a instances (hpc6a.48xlarge) with Elastic Fabric Adapter (EFA) for low-latency inter-node communication. The job is tightly coupled, requiring high network throughput. They use a placement group with cluster placement to minimize latency. A misconfiguration is using standard instances without EFA, which results in high network latency and poor scaling. The exam may test that HPC Optimized instances support EFA and are designed for tightly coupled workloads.
SAA-C03 Exam Coverage
This topic falls under Domain 2: Resilient Architectures (Objective 2.1: Choose the appropriate AWS services based on compute, storage, database, and networking requirements). The exam tests your ability to select the correct EC2 instance family for a given workload. Expect 2-3 questions directly on instance family selection, and additional questions that require knowledge of instance types as part of a larger architecture.
Common Wrong Answers and Why
Choosing General Purpose for CPU-intensive workloads: Candidates often pick M5 because it's "balanced." The correct answer is Compute Optimized (C5/C6g) because CPU is the bottleneck. The exam may describe a batch processing job with high CPU usage; M5 would work but is more expensive and less performant than C5.
Choosing Compute Optimized for memory-intensive workloads: If the question mentions "large in-memory cache," candidates may think compute is needed for speed. The correct answer is Memory Optimized (R5/X1) because memory capacity is the primary requirement.
Using T instances for databases: T instances are burstable and not suitable for sustained loads. Databases require consistent I/O and CPU. The correct answer is M5 or R5. The exam may ask about a production database with steady state load; T instances would exhaust credits.
Selecting Storage Optimized for general storage needs: Storage Optimized instances are for high I/O, not for large storage capacity. For large storage, use EBS on a General Purpose instance. The exam may describe a file server needing 10 TB of storage; the correct answer is an M5 instance with EBS, not an I3 instance.
Specific Numbers and Terms
Burstable credits: T3 instances earn 6 credits per vCPU per hour, max credit balance 144 credits (for t3.nano). The baseline CPU utilization is 10% for t3.nano (1 vCPU).
Naming: t3.micro = 1 vCPU, 1 GiB memory; c5.xlarge = 4 vCPU, 8 GiB memory; r5.xlarge = 4 vCPU, 32 GiB memory.
EFA: Elastic Fabric Adapter for HPC, supports up to 100 Gbps.
Graviton: AWS ARM-based processors (A1, T4g, M6g, C6g, R6g, etc.) offering up to 40% better price-performance.
Instance store: NVMe SSD on I3, D2, etc. Data is ephemeral.
Edge Cases
Nitro instances: All current generation instances (5th gen and later) use the Nitro hypervisor, which provides better performance and security.
Bare metal instances: Some families (e.g., m5.metal, i3.metal) provide direct access to the physical hardware for workloads that require it (e.g., licensing).
Mac instances: For macOS development, use mac1.metal.
High Memory instances: u-6tb1.metal, u-9tb1.metal, u-12tb1.metal for SAP HANA.
How to Eliminate Wrong Answers
If the workload is described as "CPU-intensive" or "high compute," eliminate any family that is not Compute Optimized.
If the workload requires large amounts of RAM (e.g., in-memory database), eliminate families with low memory-to-vCPU ratios.
If the workload requires high disk I/O (e.g., large NoSQL database), consider Storage Optimized, but only if the data can be ephemeral; otherwise, use EBS with a General Purpose instance.
If the workload uses GPUs, eliminate all non-Accelerated Computing families.
If the workload is variable and cost-sensitive, consider T family but ensure sustained load is within baseline.
EC2 instance families are General Purpose (T, M), Compute Optimized (C), Memory Optimized (R, X), Accelerated Computing (P, G, Inf, Trn), Storage Optimized (I, D), and HPC Optimized (Hpc).
The naming convention is [family][generation].[size], e.g., t3.micro, c5.xlarge.
Burstable T instances earn CPU credits at 6 credits per vCPU per hour; max credit balance is 144 credits for t3.nano.
Compute Optimized instances have a high CPU-to-memory ratio (1:2) and high clock speeds (up to 3.6 GHz).
Memory Optimized instances have a high memory-to-CPU ratio (1:8 or more), ideal for in-memory databases.
Storage Optimized instances provide high I/O with NVMe instance store, but data is ephemeral.
Accelerated Computing instances include GPUs (P3, P4, G4, G5) or AWS Inferentia (Inf1) for ML and graphics.
HPC Optimized instances support Elastic Fabric Adapter (EFA) for low-latency inter-node communication.
Instance type can be changed after stop (for most types) using modify-instance-attribute.
Always match the instance family to the workload's primary bottleneck: CPU, memory, storage, or network.
These come up on the exam all the time. Here's how to tell them apart.
General Purpose (M5)
Balanced CPU, memory, and networking.
Suitable for web servers, small databases, development environments.
Lower CPU clock speed (up to 3.1 GHz).
Memory-to-vCPU ratio typically 1:4.
Cost per hour: m5.large ~$0.096.
Compute Optimized (C5)
High CPU-to-memory ratio (1:2).
Suitable for batch processing, video encoding, scientific modeling.
Higher CPU clock speed (up to 3.6 GHz).
Lower memory per vCPU.
Cost per hour: c5.large ~$0.085.
Memory Optimized (R5)
High memory-to-vCPU ratio (1:8 or higher).
Suitable for in-memory databases (Redis, Memcached), real-time analytics.
Uses EBS for persistent storage.
Network bandwidth up to 25 Gbps.
Cost per hour: r5.large ~$0.126.
Storage Optimized (I3)
High I/O performance with NVMe instance store.
Suitable for NoSQL databases (Cassandra, MongoDB), data warehousing.
Instance store is ephemeral; data lost on stop/terminate.
Network bandwidth up to 25 Gbps.
Cost per hour: i3.large ~$0.156.
Burstable (T3)
CPU credits allow bursting above baseline.
Cost-effective for variable workloads with low average CPU.
Baseline CPU utilization: 10% for t3.nano (1 vCPU).
Not suitable for sustained high CPU; credits exhaust.
Cost per hour: t3.micro ~$0.0104.
Standard (M5)
Consistent CPU performance regardless of load.
Suitable for steady-state workloads.
No credit mechanism; always full performance.
Higher cost for low-utilization workloads.
Cost per hour: m5.large ~$0.096.
Mistake
All EC2 instances have the same CPU performance per vCPU.
Correct
CPU performance varies by family and generation. Compute Optimized instances have higher clock speeds (e.g., up to 3.6 GHz) and use different processors (Intel Xeon, AMD EPYC, AWS Graviton). General Purpose instances may have lower baseline performance. Burstable instances (T3) have a baseline that depends on credit balance.
Mistake
Burstable instances (T family) are suitable for any workload because they can burst.
Correct
Burstable instances are only suitable for workloads with low baseline CPU and occasional spikes. Sustained high CPU usage depletes credits, causing performance throttling. For steady-state workloads, use standard instances (M, C, R).
Mistake
Instance store volumes are persistent across stop/start.
Correct
Instance store volumes are ephemeral. Data is lost when the instance is stopped, terminated, or fails. They are suitable for temporary data like caches, scratch space, or replicated data. For persistent storage, use EBS.
Mistake
Larger instance sizes always provide better performance per dollar.
Correct
Larger instances often have better network and EBS bandwidth, but they are more expensive. For workloads that scale horizontally (e.g., web servers), many smaller instances may be more cost-effective. The exam tests cost optimization by comparing instance sizes.
Mistake
All instance families support the same EBS volume types and maximum IOPS.
Correct
EBS performance limits vary by instance family and size. For example, t3.nano has a maximum EBS bandwidth of 0.208 Gbps, while t3.2xlarge has 4.5 Gbps. Provisioned IOPS volumes (io1/io2) require instances with EBS-optimized capability, which is available on most current generation instances but not on all sizes.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
T3 instances are burstable, meaning they accumulate CPU credits when idle and consume them when active. They are cost-effective for workloads with low average CPU utilization and occasional spikes. M5 instances provide consistent, full CPU performance at all times, making them suitable for steady-state workloads. T3 instances have a baseline CPU performance (e.g., 10% for t3.nano) and can burst above that until credits are exhausted. M5 instances have no credit mechanism. For exam questions, if a workload has variable CPU usage with occasional spikes, choose T3. If it requires sustained high CPU, choose M5.
It depends on the database type. If the database is CPU-bound (e.g., high query processing), a Compute Optimized instance (C5) may be appropriate. However, most databases are memory-bound (e.g., caching) or I/O-bound. For in-memory databases like Redis, use Memory Optimized (R5). For transactional databases like MySQL, General Purpose (M5) or Memory Optimized may be better. The exam will provide clues about the bottleneck. If the question says "high CPU usage during queries," then Compute Optimized is correct.
When CPU credits are exhausted, the instance is throttled to its baseline CPU performance. For example, a t3.micro has a baseline of 10% (1 vCPU at 10% = 0.1 vCPU). This can cause significant performance degradation. To avoid this, you can enable T2 Unlimited (T3 is always unlimited by default) which allows the instance to burst above baseline even when credits are zero, but you incur additional charges. The exam may test that T3 Unlimited is the default and charges apply for surplus credits.
For machine learning training, Accelerated Computing instances with GPUs are best, such as P3 (NVIDIA Tesla V100), P4 (NVIDIA A100), or Trn1 (AWS Trainium). These provide the parallel processing power needed for deep learning. For inference, Inf1 (AWS Inferentia) offers lower cost. Do not use Compute Optimized or General Purpose for GPU workloads. The exam may ask about training vs. inference; training requires P3/P4, inference can use Inf1.
Instance store provides temporary, block-level storage directly attached to the host computer. It offers high I/O performance but data is lost when the instance is stopped or terminated. EBS provides persistent, network-attached storage that survives instance stop/terminate. Instance store is suitable for caches, scratch data, and replicated data. EBS is for persistent data like databases. The exam often tests that instance store is ephemeral and should not be used for critical data.
For most instance types, you must stop the instance before changing its type. After stopping, you can use the AWS Console, CLI, or SDK to modify the instance type. Some instances (e.g., T2) do not support changing to certain types without stopping. Additionally, the new type must be compatible with the instance's network and storage configuration. The exam may test that you cannot change the instance type while it is running.
AWS Graviton instances (e.g., T4g, M6g, C6g, R6g) use ARM-based processors designed by AWS. They offer up to 40% better price-performance compared to x86 instances for many workloads. They are ideal for web servers, containerized microservices, and open-source databases. However, some software may not be compatible with ARM architecture. The exam may ask about Graviton for cost optimization in a scenario where the application is compatible.
You've just covered EC2 Instance Type Families and Use Cases — now see how well it sticks with free SAA-C03 practice questions. Full explanations included, no account needed.
Done with this chapter?