When an organisation needs network storage, the question is not just how much space is required, but how applications will access that data. There are three fundamentally different approaches to solving this problem: iSCSI operates at the block level, NFS at the file level, and Object Storage at the object level. Each one has clear strengths, specific limitations, and use cases where it is the optimal choice.
Choosing the wrong protocol can result in lost performance, infrastructure overspend or scalability limitations that only become apparent as the environment grows. Choosing well means aligning the access pattern of each workload with the protocol that best serves it, and in many cases, the right answer is to combine all three.
In this article we explain how each protocol works, compare their performance characteristics, detail their primary use cases and show how EasyDataHost offers all three storage models to cover any infrastructure need.
iSCSI: Block Storage over TCP/IP
iSCSI (Internet Small Computer Systems Interface) is a protocol that transports SCSI commands over standard TCP/IP networks. In practical terms, it allows a server to access a remote disk as if it were a local disk connected directly via SATA or SAS. The operating system sees a raw block device, with no awareness that the data travels over the network.
The iSCSI architecture is based on two roles: the initiator (the client requesting disk access) and the target (the server or array presenting the storage volumes). Volumes are exposed as LUNs (Logical Unit Numbers), and each LUN behaves like an independent disk that the operating system can partition, format and mount with any file system.
The main advantage of iSCSI is that it inherits all the semantics of local disks: it supports any file system, works with databases that need direct block-level I/O, allows volume-level snapshots and is compatible with the HA mechanisms of hypervisors such as Proxmox or Hyper-V. Because it runs over standard TCP/IP, it does not require specialised hardware like Fibre Channel, making it a low-cost SAN alternative.
NFS: Shared Network File System
NFS (Network File System) is a file-level protocol that allows remote directories to be mounted as if they were local. Unlike iSCSI, which exposes raw blocks, NFS presents a complete file system with directories, POSIX permissions, locks and metadata. Multiple clients can mount the same NFS export simultaneously and access the same files.
NFS has evolved significantly since its origins. NFSv3 is still widely used for its simplicity and compatibility, but it operates statelessly and relies on auxiliary protocols for file locking. NFSv4.x integrates locking natively, supports Kerberos authentication, operates over a single TCP port (which simplifies firewalls) and improves performance with mechanisms such as delegations and compound sessions.
The POSIX semantics of NFS make it ideal for workloads that require shared file access: multiple web servers serving the same content, video editing workflows where several editors access the same assets, shared home directories in corporate environments, or persistent volumes for containers. Synology NAS appliances are a classic example of an NFS server in SMB environments.
Object Storage: Flat Namespace with HTTP/S3 API
Object Storage breaks away from the traditional block and file model. Instead of directories and hierarchical paths, data is stored as objects in a flat namespace. Each object consists of three elements: the data itself, a unique identifier (key) and arbitrary metadata describing the content. Access is performed through an HTTP API, typically S3-compatible.
This architecture has profound implications for scalability. By not maintaining a directory hierarchy or complex metadata tables, Object Storage can scale to exabytes of data without degradation in indexing performance. Objects are immutable by default (they are created, read and deleted, but not partially modified), which simplifies replication, versioning and consistency management in distributed systems.
The trade-off is that Object Storage does not support typical file system operations: you cannot mount a bucket as a disk, you cannot perform random seeks within an object, and you cannot use POSIX locks. Per-operation latency is higher than iSCSI or NFS because each access is a full HTTP request. You can learn more about scalability advantages in our article on object storage and scalability.
Key concept:
iSCSI presents blocks (like a local disk), NFS presents files (like a shared folder) and Object Storage presents objects (via HTTP API). They are not direct competitors: each one addresses a different access pattern.
Comparison Table: iSCSI vs NFS vs Object Storage
The following table summarises the fundamental differences between the three protocols across the aspects that matter most when making a choice:
| Criterion | iSCSI | NFS | Object Storage (S3) |
|---|---|---|---|
| Protocol layer | Block (SCSI over TCP/IP) | File (RPC/NFS over TCP) | Object (HTTP/HTTPS REST API) |
| Access pattern | Random block read/write | Shared file read/write | PUT/GET of whole objects |
| Performance | Lowest latency, highest IOPS | Good for sequential/shared access | High throughput, higher per-operation latency |
| Scalability | Limited by target/LUN | Limited by NFS server | Practically unlimited (exabytes) |
| Primary use case | VM disks, databases, SAN | Shared files, web, containers | Backup, archiving, data lakes, CDN |
| Multitenancy | Typically one LUN per client | Exports per client, POSIX permissions | Native (buckets, IAM policies, ACLs) |
| Cost per TB | High (requires fast storage) | Medium | Low (commodity hardware, erasure coding) |
Performance Characteristics
Each protocol's performance responds to fundamentally different access patterns, and comparing them with a single metric is a common mistake. Each has its own performance profile that makes it optimal for specific workloads:
- speed iSCSI: delivers the lowest latency and highest IOPS for random block operations. By transporting native SCSI commands, protocol overhead is minimal. With NVMe disks on the target and 10/25 GbE networking, iSCSI achieves latencies below 200 microseconds and hundreds of thousands of IOPS. It is the protocol of choice when random I/O performance is critical.
- share NFS: excels at sequential access and shared workloads. Protocol overhead is higher than iSCSI (because it manages files, metadata and locks), but NFSv4.x with delegations and compound sessions significantly reduces metadata latency. For sequential reads/writes of large files, NFS makes good use of available network bandwidth.
- cloud_upload Object Storage: optimised for high throughput with large objects. Each operation is a full HTTP request, which adds latency compared with iSCSI or NFS (typically tens of milliseconds). However, it supports massive parallelism: hundreds of clients can perform PUT/GET operations simultaneously without contention, and multipart uploads can saturate available bandwidth with objects of any size.
iSCSI Use Cases
iSCSI is the protocol of choice when applications need a raw block device with the lowest possible latency:
- dns Virtual machine disks: hypervisors such as Proxmox, Hyper-V and VMware ESXi support iSCSI datastores. Each VM sees a block disk that can be formatted with ext4, XFS or NTFS. Ideal for the EasyDataHost Cloud IaaS platform.
- database Transactional databases: engines such as PostgreSQL, MySQL/MariaDB and SQL Server need block-level access with low-latency random I/O. iSCSI provides them with a remote disk that has identical semantics to a local disk, without the overhead of an intermediate file layer.
- lan Fibre Channel SAN replacement: iSCSI over 10/25 GbE Ethernet networks delivers comparable performance to FC at a fraction of the cost, eliminating the need for dedicated FC switches and specialised HBAs.
NFS Use Cases
NFS is the natural protocol when multiple servers need concurrent access to the same files:
- folder_shared Shared home directories: in corporate and academic environments, NFS allows users to access their files from any workstation. POSIX permissions ensure isolation between users.
- videocam Media workflows: video editing, 3D rendering and post-production require multiple editors and render nodes to access the same assets simultaneously. NFS with sufficient bandwidth supports these high sequential throughput workflows.
- language Shared web content: web server farms serving the same static files can mount a common NFS export instead of synchronising files between nodes.
- deployed_code Container persistent volumes: Kubernetes supports NFS as a backend for PersistentVolumes, allowing pods on different nodes to share data. VMware ESXi also supports NFS datastores as an alternative to iSCSI.
Object Storage Use Cases
Object Storage with an S3 API is the optimal choice when scalability, cost per TB and HTTP accessibility are the priority:
- backup Backup target (Veeam + S3): tools such as Veeam Backup & Replication support S3 repositories natively, including immutability (object lock) for ransomware protection. EasyDataHost S3 storage is an ideal destination for offsite backups.
- analytics Data lakes and analytics: frameworks such as Apache Spark, Presto and Trino consume data directly from S3 buckets. The flat namespace and HTTP API simplify large-scale data ingestion from multiple sources.
- perm_media Media archives and CDN origin: images, videos, documents and static assets are stored as objects and served directly via HTTP or through a CDN. The metadata associated with each object facilitates cataloguing and search.
- inventory_2 Application data: web and mobile applications use the S3 protocol as a native storage layer for user uploads, logs, exports and any unstructured data that does not fit in a relational database.
Hybrid Approach: Combining All Three Protocols
The reality of most production infrastructures is that they need all three storage types simultaneously. It is not about choosing one and discarding the others, but about assigning each workload to the protocol that best fits its access pattern:
- view_in_ar iSCSI for VM disks and databases: where minimal latency and IOPS are critical. Hypervisors access iSCSI LUNs as primary storage for production virtual machines.
- folder_shared NFS for shared files: web content, media assets, project directories and Kubernetes volumes where multiple nodes need concurrent access to the same file system.
- cloud_upload S3 for backup, archiving and bulk data: Veeam repositories, data lakes, historical archives and any data accessed via HTTP API. The lowest cost per TB and unlimited scalability make S3 the natural choice for high-volume storage.
Platforms such as Ceph unify all three access models (RBD for blocks, CephFS for files, RGW for objects) on a single distributed infrastructure, simplifying management and reducing the number of systems to maintain.
Practical advantage:
Do not choose a single protocol for your entire infrastructure. Use iSCSI where you need IOPS and low latency, NFS where you need shared files, and S3 where you need massive scalability and low cost. The key is to align each workload with the right protocol.
Storage Options at EasyDataHost
EasyDataHost offers all three network storage models to cover any infrastructure need, all with data hosted in a Tier III+ data centre in Madrid:
- check_circle S3 Storage: S3 API-compatible, ideal for backup with Veeam, data lakes, archiving and application data. Scales without practical limits using erasure coding.
- check_circle Storage servers: dedicated servers with large disk capacities (HDD and NVMe) that you can configure as iSCSI targets, NFS servers or nodes in a Ceph cluster.
- check_circle Cloud IaaS: virtual machines with NVMe Ceph storage in triple replication. High-performance blocks for production VMs with live migration and high availability.
- check_circle Managed services: design, implementation and maintenance of hybrid storage architectures: iSCSI target configuration, NFS exports, S3 buckets and Ceph clusters.
Conclusion
iSCSI, NFS and Object Storage are not interchangeable alternatives: they are complementary protocols that address fundamentally different access patterns. Understanding the strengths of each one is the foundation for building a storage architecture that is efficient, scalable and aligned with the real needs of each workload.
- arrow_right iSCSI is block storage over TCP/IP: lowest latency, highest IOPS, ideal for VMs and databases.
- arrow_right NFS is a shared network file system: concurrent access, POSIX semantics, ideal for shared files and containers.
- arrow_right Object Storage is a flat namespace with an S3 API: massive scalability, low cost per TB, ideal for backup, archiving and data lakes.
- arrow_right The hybrid approach combines all three: iSCSI for IOPS, NFS for shared files, S3 for bulk volume.
- arrow_right EasyDataHost offers S3, storage servers, Cloud with NVMe Ceph and managed services to cover all three models.
If you need help designing the storage architecture for your infrastructure, contact our team to analyse your workloads and recommend the right combination of protocols.