S3 (Simple Storage Service) for object storage + EFS (Elastic File System) for shared file storage. Complete guide with real Indian company scenarios.
Amazon S3 (Simple Storage Service) is AWS's object storage service. Unlike EBS (which is a hard drive attached to one EC2), S3 is an internet-accessible storage — you can store any file and access it via a URL from anywhere in the world.
S3 is one of AWS's oldest and most-used services. It stores objects (files + metadata) inside buckets (containers). There is no folder structure in S3 — only the illusion of folders through "/" in object names.
You upload files. Each file gets a URL. Any application (or person) with permission can download it via that URL. Unlike Google Drive for humans, S3 stores billions of files and serves billions of requests per day — at any scale. It never gets "full" — storage is unlimited, you only pay for what you use.
No capacity limit. Store 1 file or 1 billion files. Individual objects up to 5 TB.
Every object has a unique URL. Access it from any browser, app, or region worldwide.
99.999999999% — AWS replicates your data across 3+ AZs automatically. Almost impossible to lose data.
Pay only for storage used + data transferred. No upfront commitment. Prices from $0.023/GB/month.
Hotstar (now Disney+ Hotstar) streams to 500 million+ users. Every cricket match, movie, and web series is stored on AWS S3. The video files (each season of a show can be 500 GB+), thumbnail images, subtitle files, and chapter markers are all stored as S3 objects. When you open Hotstar and see a show thumbnail — that image is being served from an S3 bucket via Amazon CloudFront (CDN). S3's 11-nines durability ensures Hotstar has never lost a single video file since launch.
Zepto (10-minute grocery delivery app) stores all product images — photos of 10,000+ grocery items — in S3. Each product has 3-4 high-resolution images. S3 serves these images to the Zepto app and website. When a Zepto engineer updates a product image, they just replace the S3 object — the app everywhere automatically shows the new image within minutes.
S3 has a simple two-level structure: Buckets contain Objects.
images/products/zepto-apple.jpg).| Concept | Details | Example |
|---|---|---|
| Bucket Name | Globally unique, 3-63 chars, lowercase, no underscores | zomato-food-images-prod |
| Object Key | Full path to object inside bucket | restaurants/delhi/biryani.jpg |
| Object URL | https://[bucket].s3.[region].amazonaws.com/[key] | https://zomato-food-images-prod.s3.ap-south-1.amazonaws.com/restaurants/delhi/biryani.jpg |
| Max Object Size | 5 TB per single object (use multipart upload for >100 MB) | A 2-hour HD movie = ~4 GB — fine for S3 |
| Bucket Region | Chosen at creation, cannot be changed | Mumbai (ap-south-1) for Indian user data |
Zomato organizes its S3 buckets like this:
• zomato-restaurant-images-prod → Food photos uploaded by restaurants (1M+ images)
• zomato-user-profile-photos → Profile pictures of 100M+ users
• zomato-order-invoices-2026 → PDF invoices for all orders (compliance)
• zomato-analytics-raw-data → Click-stream data for data science team
Inside zomato-restaurant-images-prod, objects are organized: cities/delhi/restaurants/biryani-house/dish-1.jpg. This isn't a real folder — it's just "/" in the key name, but the S3 Console shows it as folders for convenience.
S3 has 7 storage classes — you pay less for data you access less frequently. Choosing the right class saves significant money. All classes provide 11-nines durability.
Highest availability (99.99%). Instant access. Replicated across 3+ AZs. For active production data.
AWS automatically moves objects between Frequent and Infrequent tiers. Small monitoring fee per 1,000 objects.
Lower storage cost, retrieval fee charged. Good for data accessed once a month or less. Minimum 30-day storage.
Stored in only ONE AZ (not 3). 20% cheaper than Standard-IA. OK if you can recreate the data if that AZ fails.
For archive data accessed once per quarter. Millisecond retrieval time. Minimum 90 days storage.
Retrieval in 1 min–12 hours. For data accessed once or twice a year. Minimum 90 days storage.
Retrieval within 12 hours. Lowest cost storage in all of AWS. For data you almost never access — 7-10 year archive. Minimum 180 days.
BookMyShow stores current movie showtimes, seat maps, and booking confirmations in S3 Standard — accessed millions of times per day. But old ticket PDFs from 2020–2023 (accessed maybe once a year when customers need a refund) are stored in Glacier Flexible Retrieval — 85% cheaper. BookMyShow saves ₹40+ lakhs per year in storage costs just by choosing the right class for different data ages.
IRCTC must retain all railway booking records for 7 years (government compliance). Records older than 2 years are automatically moved to Glacier Deep Archive — at ₹0.066/GB/month ($0.00099) — the cheapest possible storage. 7 years of booking data at 50 TB = costs only ₹3,300/month vs ₹1.15 lakhs/month on S3 Standard. That's 97% savings.
By default, S3 buckets are 100% private — no one can access anything. You control access through 3 mechanisms.
JSON policies attached to the bucket. Grant access to specific IAM users, roles, or entire public. Most flexible option.
Legacy per-object permissions. Grant read/write to specific AWS accounts. Rarely used today — Bucket Policies are preferred.
Account-level override that blocks ALL public access regardless of bucket policies. Enable this on all buckets that don't need to be public.
When you want everyone to read files (e.g., website images). Make bucket public + add bucket policy allowing public GetObject.
A Pre-signed URL is a temporary URL that gives anyone who has it time-limited access to a specific S3 object — without making the bucket public.
It is generated by your server using your AWS credentials. The URL includes an expiry time (e.g., 15 minutes, 1 hour). After expiry, the URL stops working.
The hotel (S3) is private — you can't just walk in. But the receptionist (your server) generates a special key card (pre-signed URL) that lets you enter your specific room (object) for exactly 3 days. After checkout (expiry), the key card stops working. You never gave permanent access — just a time-limited pass.
Razorpay (India's top payment gateway) generates payment receipts as PDF files stored in a private S3 bucket. When a merchant downloads a receipt, Razorpay's server generates a pre-signed URL valid for 5 minutes. The merchant clicks the link and downloads the PDF. After 5 minutes, the link expires — even if they shared it with someone, the URL no longer works. This keeps financial documents secure while allowing legitimate access.
Nykaa (beauty e-commerce) stores product demonstration videos privately on S3. When a user opens a product page, the backend generates a pre-signed URL valid for 2 hours. The video player uses this URL to stream the video. Even if a user copies the URL and tries to share it, after 2 hours it stops working. This prevents unauthorized video sharing/downloading.
Versioning keeps multiple versions of every object in the same bucket. When you upload a new version, the old version is kept — not deleted. Every version gets a unique version ID.
Accidentally deleted a file? Overwrote it with wrong data? Restore any previous version instantly.
Deleting an object adds a "delete marker" — original versions still exist. Remove the marker to restore.
Every version stored = billed separately. Use lifecycle policies to delete old versions after N days.
S3 can serve static websites (HTML, CSS, JS files — no server-side code) directly. No EC2, no web server needed. Just upload your files and enable "Static Website Hosting" in bucket settings.
Your website URL becomes: http://[bucket-name].s3-website.[region].amazonaws.com
For custom domains (like yourcompany.com), use with Amazon CloudFront (CDN) for HTTPS + speed.
Meesho hosts its seller documentation website (how-to guides, API docs, policy pages) on S3 Static Website Hosting. The entire site is HTML/CSS/JS files — no backend needed. Hosting on S3 costs ₹500–2,000/month vs running an EC2 (₹3,000–10,000/month). For any website that doesn't need a database or server-side processing, S3 is the cheapest and most reliable option.
Thousands of Indian startups host their product landing pages, coming-soon pages, and marketing microsites on S3 + CloudFront. Zero server management, infinite scalability, and near-zero cost for low-traffic sites.
Lifecycle Policies automatically transition objects to cheaper storage classes or delete them after a set number of days — without any manual intervention. Define rules once; AWS handles it forever.
Fresh object — user accesses it frequently
Lifecycle rule kicks in — infrequent access after 1 month
Rarely accessed — archive it
If data is no longer needed after 1 year
OYO Rooms collects hotel check-in photos, maintenance reports, and guest feedback documents. They use lifecycle policies: photos from last 30 days stay in S3 Standard (operations team checks them daily). Photos 30–90 days old move to Standard-IA (referenced occasionally for disputes). Photos older than 90 days move to Glacier (kept 3 years for legal compliance). After 3 years, auto-deleted. OYO saves ₹25+ lakhs per year compared to keeping everything in S3 Standard forever.
Swiggy's delivery partner app generates location GPS logs. Recent 7 days in S3 Standard (real-time analysis). 7–30 days in Standard-IA (monthly reporting). 30 days+ in Glacier Deep Archive (compliance). After 2 years, auto-deleted. This is a standard data lake pattern used by all major Indian tech companies.
S3 Replication automatically copies objects from one bucket to another — in the same region or a different region. Two types:
S3 offers multiple encryption options. Since Nov 2023, all new S3 objects are encrypted by default with SSE-S3 — no extra configuration needed.
| Type | Managed By | Best For |
|---|---|---|
| SSE-S3 (default) | AWS manages keys automatically | Standard data — zero effort |
| SSE-KMS | AWS KMS — you control key rotation and access logs | Compliance data — audit who accessed keys |
| SSE-C | You provide your own key with every request | Strict data sovereignty — key never leaves your control |
| Client-Side | You encrypt before uploading | Maximum security — encrypted even before reaching AWS |
| Concept | Amazon S3 | Azure Blob Storage |
|---|---|---|
| Service Name | Amazon S3 | Azure Blob Storage |
| Container type | Bucket | Container (inside Storage Account) |
| Object name | Object (with Key) | Blob |
| Hot storage | S3 Standard | Hot tier |
| Infrequent access | S3 Standard-IA | Cool tier |
| Archive | Glacier Deep Archive | Archive tier |
| Static Website | S3 Static Website Hosting | Azure Static Web Apps / $web container |
| Versioning | S3 Versioning | Blob Versioning (same concept) |
| CDN | Amazon CloudFront | Azure CDN / Front Door |
| Temp URLs | Pre-signed URLs | SAS (Shared Access Signature) URLs |
| Data transfer tool | AWS Snowball | Azure Data Box |
| File Storage | Amazon EFS (NFS, Linux) | Azure Files (SMB + NFS) |
Elastic File System — Shared Network File Storage for Linux EC2 instances. Multiple EC2s read/write the same file system simultaneously.
Amazon EFS is a fully managed, shared NFS (Network File System) for Linux-based EC2 instances. Unlike EBS (one volume → one EC2), EFS allows multiple EC2 instances to read and write the same file system simultaneously.
EFS is elastic — it grows and shrinks automatically as you add or remove files. You never provision capacity; you only pay for what you store.
Think of EFS like a company's shared network folder that all employees (EC2 instances) can access at the same time. Employee A can edit a document, Employee B can open the same folder and see the change immediately — they're all connected to the same storage. EBS would be like each employee having their own private hard drive — no sharing. S3 would be like a shared Google Drive — accessible via URL but not mountable as a drive letter.
Thousands of EC2 instances can mount and access the same EFS file system simultaneously.
EFS uses NFS protocol — only compatible with Linux EC2 instances. For Windows, use Amazon FSx for Windows.
Starts at 0 bytes. Grows automatically as you add files. Shrinks when you delete them. Never run out of space.
EFS file system is accessible from multiple Availability Zones in the same region. EBS is locked to a single AZ.
| Feature | EFS (File Storage) | EBS (Block Storage) | S3 (Object Storage) |
|---|---|---|---|
| Type | Shared network file system | Block device (virtual hard drive) | Object store (URL-accessed) |
| Protocol | NFS (Network File System) | Block (transparent to OS) | HTTP REST API |
| Access | Thousands of EC2s simultaneously | One EC2 at a time (except io2 multi-attach) | Anyone with URL/permission |
| OS | Linux only (NFS) | Linux and Windows | OS-independent |
| Pricing | ~$0.30/GB/month (Standard) | ~$0.08–0.10/GB/month (gp3) | ~$0.023/GB/month (Standard) |
| Capacity | Auto-elastic (petabyte scale) | Provisioned size (max 64 TB) | Unlimited |
| Use for | Shared configs, ML models, home dirs | OS drive, databases, apps | Images, videos, backups, logs |
A major OTT platform like Hotstar or SonyLIV runs 50+ video transcoding servers simultaneously — taking one uploaded video and converting it into 10 different resolutions (240p, 360p, 480p, 720p, 1080p, 4K, etc.). All 50 servers read the same original video file and write their output to the same shared folder. This is a classic EFS use case — the file system is mounted on all 50 EC2s, and they all work in parallel on shared storage.
Swiggy's ML team trains machine learning models (delivery time prediction, demand forecasting) using Python on 20 EC2 GPU instances. All 20 servers need access to the same 500 GB training dataset. With EFS, they mount the same file system on all 20 instances. With EBS, they'd need to copy 500 GB to each machine — 10 TB total, slow and expensive. EFS = one copy, all servers access it.
Freshdesk (Freshworks product) runs on a fleet of 100+ application servers. All servers need access to the same configuration files, shared libraries, and log directories. EFS is mounted at /app/config and /app/logs — when DevOps updates a config file, all 100 servers see the change immediately without any restart or sync process.
EFS offers two performance modes. Choose based on how many files per second your workload requires. This cannot be changed after creation — choose carefully.
Lowest latency per file operation. Best for most workloads. Up to 35,000 IOPS. Suitable for web servers, content management, home directories, dev environments.
Highest aggregate throughput and IOPS — but slightly higher latency per operation. For workloads with thousands of EC2s all accessing simultaneously. Big data, media processing, analytics.
Similar to S3, EFS has tiered pricing. Files you access less frequently should be in cheaper tiers. Use EFS Intelligent-Tiering to automatically move files between tiers based on access.
| Storage Class | Price (approx.) | Use For |
|---|---|---|
| EFS Standard | ~$0.30/GB/month | Actively accessed files (hot data) |
| EFS Standard-Infrequent Access | ~$0.025/GB/month + retrieval fee | Files not accessed for 30+ days |
| EFS One Zone | ~$0.16/GB/month | Single AZ (cheaper, less resilient) — dev/test workloads |
| EFS One Zone-Infrequent Access | ~$0.013/GB/month + retrieval fee | Single AZ + rarely accessed — cheapest EFS option |
This is the most important table in this module. Know this cold — exam questions always compare these three.
| Question | Use EBS 💾 | Use S3 🗂️ | Use EFS 📁 |
|---|---|---|---|
| What do I store? | OS disk, database files, app binaries | Images, videos, backups, logs, documents | Shared configs, shared datasets, home directories |
| How many EC2s access it? | 1 (exclusive) | ∞ (via URL) | Thousands simultaneously |
| Can it run a database? | ✅ Yes — fast IOPS | ❌ No — not a file system | ⚠️ Possible but not recommended |
| Works on Windows? | ✅ Yes | ✅ Yes (REST API) | ❌ No — NFS = Linux only |
| Static website hosting? | ❌ No | ✅ Yes | ❌ No |
| Storage grows automatically? | ❌ No — manually provisioned | ✅ Yes — unlimited | ✅ Yes — elastic |
| AWS Pricing (approx.) | $0.08–0.10/GB/mo (gp3) | $0.023/GB/mo (Standard) | $0.30/GB/mo (Standard) |
| Azure equivalent | Azure Managed Disks | Azure Blob Storage | Azure Files |
| Service | Topic | AWS Docs URL |
|---|---|---|
| S3 | S3 Overview | S3/Welcome.html |
| S3 | Storage Classes | storage-class-intro.html |
| S3 | Pre-signed URLs | ShareObjectPreSignedURL.html |
| S3 | Lifecycle Policies | object-lifecycle-mgmt.html |
| EFS | EFS Overview | efs/whatisefs.html |
| EFS | Performance Modes | efs/performance.html |
| EFS | EFS Storage Classes | efs/storage-classes.html |