Object vs Block vs File Storage: Fundamentals & Selection

Cloud infrastructure provides three fundamental storage abstractions, each engineered for distinct latency, access patterns, and scalability requirements:

1. Block Storage (AWS EBS, Google Persistent Disk, Local NVMe):

• Exposes raw, unformatted storage sectors (e.g., 4KB or 8KB blocks) directly to the operating system over high-speed protocols (NVMe-oF, iSCSI).
• The operating system formats the raw block device with a filesystem (ext4, XFS, NTFS).
• Provides the lowest latency (< 1 ms) and highest random I/O performance (up to 256,000 IOPS on AWS io2 Block Express).
• Limitation: Can typically only be attached to a single virtual machine or pod at a time (ReadWriteOnce).

2. File Storage (AWS EFS, Azure Files, NFS, SMB/CIFS):

• Presents a traditional hierarchical directory tree (/shared/uploads/2026/01/invoice.pdf) accessible over the network via NFSv4 or SMB.
• Supports standard POSIX filesystem semantics, directory traversal, and distributed file locks.
• Advantage: Supports concurrent read and write access from hundreds or thousands of instances simultaneously (ReadWriteMany).
• Limitation: Higher latency (3 - 15 ms) and significantly higher cost per gigabyte (sim \0.30/GB-month$).

3. Object Storage (Amazon S3, Google Cloud Storage, Azure Blob, MinIO):

• Stores data as distinct, immutable Objects inside a flat, non-hierarchical namespace (Bucket + Object Key).
• Each object contains the raw binary payload, a unique key, and arbitrary customizable metadata headers.
• Accessed entirely over HTTP/HTTPS REST APIs (GET, PUT, DELETE, HEAD).
• Advantage: Infinite horizontal scale, 11 Nines (99.999999999\%) of durability, built-in multi-AZ redundancy, and lowest cost (sim \0.023/GB-month$).
• Limitation: High initial first-byte latency (20 - 100 ms) and no random in-place byte updates (modifying a single byte requires replacing the entire object).