Design High Performing Architectures · Topic 1 of 7
High Performing Storage Solutions
Choose block, file and object storage on AWS for performance, comparing EBS volume types, instance store, EFS, FSx and S3 features that raise throughput.
- 5 min
- EBS is block storage in one AZ, and the volume type decides IOPS and throughput.
- gp3 is the general purpose default, and io2 is for the highest and most consistent IOPS.
- st1 and sc1 are HDD types for sequential workloads, and they cannot be boot volumes.
What it is
AWS offers three storage models. Block storage acts like a raw disk attached to one instance.
File storage is a shared file system mounted by many instances. Object storage holds objects in buckets and is accessed through an API.
Picking the right model, and then the right tier inside it, is the heart of this topic. The exam describes a workload (a database, a shared home directory, a data lake) and asks which storage meets its performance and scaling needs.
How it works
Block storage: EBS and instance store
Amazon Elastic Block Store (EBS) volumes exist in one Availability Zone, persist independently of the instance, and can be snapshotted. The volume type sets the performance.
| Type | Media | Use it for |
|---|---|---|
| gp3 (General Purpose SSD) | SSD | Most workloads. Baseline 3,000 IOPS and 125 MiB/s, and you can raise IOPS and throughput without growing the size |
| gp2 | SSD | Older general purpose type, where IOPS grows with volume size |
| io2 and io1 (Provisioned IOPS SSD) | SSD | Databases that need high and consistent IOPS. Supports Multi Attach to several instances in one AZ |
| st1 (Throughput Optimized HDD) | HDD | Large sequential reads, such as logs and big data. Cannot be a boot volume |
| sc1 (Cold HDD) | HDD | Infrequently accessed data at the lowest HDD cost. Cannot be a boot volume |
IOPS means input and output operations per second. Choose by the bottleneck: IOPS for small random reads and writes, throughput for large sequential transfers.
Instance store is disk physically attached to the host. It gives very high speed and low latency, but the data is lost when the instance stops, fails or is terminated. Use it for caches, buffers and scratch data, or for replicated data where loss is acceptable.
File storage: EFS and FSx
| Service | What it is | Pick it for |
|---|---|---|
| Amazon EFS | Managed Network File System (NFS) for Linux, elastic, spans Availability Zones | Shared Linux content, web serving, home directories |
| FSx for Windows File Server | Windows native file shares over Server Message Block (SMB), with Active Directory | Windows applications |
| FSx for Lustre | High performance parallel file system, linked to S3 | High performance computing (HPC), machine learning, fast processing of S3 data |
| FSx for NetApp ONTAP | Multi protocol NetApp features | Migrating NetApp workloads, NFS, SMB and iSCSI together |
| FSx for OpenZFS | ZFS based file system | Migrating ZFS workloads |
EFS grows and shrinks automatically. It has a throughput mode setting: bursting, provisioned, or elastic, which scales throughput with demand. For the exam, EFS is the default answer for "shared file system for Linux across many instances".
Object storage: Amazon S3
S3 scales almost without limit and provides high request rates per prefix, so spreading keys across prefixes raises total parallel throughput. These features speed up transfers.
- Multipart upload: splits a large object into parts that upload in parallel and can be retried individually. It is recommended for large objects and required for the biggest ones.
- S3 Transfer Acceleration: uses CloudFront edge locations to speed up uploads from distant users to a bucket.
- Byte range fetches: download parts of an object in parallel, or only the part you need.
- S3 Express One Zone: a storage class for latency sensitive workloads that need very fast, frequent access in one AZ.
Matching the need
- An IOPS heavy database on EC2: io2 or gp3.
- Shared Linux files: EFS.
- Windows shares: FSx for Windows File Server.
- Fast scratch storage that can be lost: instance store.
- A data lake or static content: S3, often with CloudFront in front.
The words "shared by many instances", "POSIX" or "NFS" point to EFS. "SMB" or "Active Directory" points to FSx for Windows. "HPC" or "sub millisecond access to S3 data" points to FSx for Lustre.
What to remember for the exam
- EBS is block storage in one AZ, and the volume type decides IOPS and throughput.
- gp3 is the general purpose default, and io2 is for the highest and most consistent IOPS.
- st1 and sc1 are HDD types for sequential workloads, and they cannot be boot volumes.
- Instance store is fast but temporary, so data is lost when the instance stops or fails.
- EFS is a shared elastic file system for Linux across AZs. FSx covers Windows, Lustre, NetApp ONTAP and OpenZFS.
- S3 performance tools are multipart upload, Transfer Acceleration, byte range fetches and spreading across prefixes.
- Put CloudFront in front of S3 to speed up delivery of content to readers.
- A snapshot copy is how you move an EBS volume to another AZ or Region.
Common mistakes
Attaching one EBS volume to instances in different AZs. A volume is tied to a single AZ. For shared access across AZs, use EFS.
Storing important data on instance store. It disappears with the instance, so keep durable data on EBS, EFS or S3.
Choosing HDD for a boot volume or random I/O. HDD types suit large sequential access only.
Using EFS for Windows. Windows file shares use SMB, which is the role of FSx for Windows File Server.
Provisioned IOPS volumes cost more than general purpose volumes. If the scenario never states a demanding IOPS requirement, gp3 is usually the right and cheaper answer.
Check yourself
Which storage gives a shared file system to Linux instances in several AZs?
Amazon EFS.
What happens to instance store data when the instance is stopped?
The data is lost, because instance store is temporary storage tied to the host.
Which S3 feature speeds up uploads from users far from the bucket's Region?
S3 Transfer Acceleration, which uses CloudFront edge locations.
Topic 1 of 7 in Design High Performing Architectures
Test yourself on Design High Performing Architectures
The full AWS SAA-C03 question bank has 823 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by AWS. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.