What is iSCSI?
iSCSI is a storage protocol that encapsulates SCSI block-level commands over TCP/IP, enabling servers to access remote disks as if they were locally attached.
iSCSI (Internet Small Computer Systems Interface) is a transport protocol that carries SCSI commands over a TCP/IP network. It enables block-level storage access, meaning a server can mount a remote storage device and treat it as a direct-attached disk. This makes iSCSI a common building block for storage area networks (SANs) using standard Ethernet infrastructure.
In an iSCSI setup, the client is called the initiator and the storage device is the target. The initiator sends SCSI CDBs (Command Descriptor Blocks) inside TCP segments to the target, typically on port 3260. The target executes the commands and returns status or data. Each target exports one or more logical unit numbers (LUNs), which appear as raw block devices to the initiator. The protocol supports authentication (CHAP, SRP), header and data digests for integrity, and error recovery mechanisms to handle packet loss without corrupting the data stream.
iSCSI operates at the boundary between the SCSI layer and the TCP/IP stack. It is often deployed as a lower-cost alternative to Fibre Channel SANs because it runs on commodity Ethernet switches and network interface cards. Performance can be improved with hardware offload engines (iSCSI HBAs) or by layering iSCSI over RDMA (iSER). iSCSI remains widely used in hypervisors, storage appliances, and cloud environments where block storage must be shared across multiple hosts.
Key facts
- iSCSI uses TCP port 3260 by default for target connections.
- The protocol is defined in RFC 7143, which consolidates earlier iSCSI specifications.
- iSCSI initiators and targets are identified by iSCSI Qualified Names (IQNs) or Extended Unique Identifiers (EUIs).
- Supports multiple connections per session for load balancing and failover.
- Authentication is commonly implemented with CHAP (Challenge-Handshake Authentication Protocol).
How it works in practice
Related terms
References
More in Storage
Block Storage
Block storage is a storage architecture that presents raw fixed-size chunks (blocks) to a computer, which the operating system partitions and formats before use, commonly used for virtual machine disks and database volumes.
Cold Storage
Cold storage is a low-cost, slow-retrieval data tier for infrequently accessed information, such as backups, archives, or compliance records, with retrieval times ranging from minutes to hours.
Erasure Coding
Erasure coding is a data protection method that splits data into fragments (shards) and adds parity, allowing reconstruction from a subset of fragments; it is commonly used in cloud object storage to tolerate failures efficiently.
Fibre Channel
Fibre Channel is a high-speed networking standard primarily used to connect storage devices in storage area networks, operating as a dedicated protocol stack separate from Ethernet.
File Storage
File storage exposes data as a hierarchical filesystem over network protocols like NFS and SMB, commonly delivered via network-attached storage (NAS) appliances for shared access.
Hot Storage
Hot storage is the default high-performance tier in object storage systems, optimized for low-latency access to frequently retrieved data.
NFS
NFS (Network File System) is a distributed file system protocol originally developed by Sun Microsystems that allows a client to access files over a network as if they were local. It is widely used in Unix and Linux environments and runs over IP networks.
Object Storage
Object storage is a flat data storage architecture that manages data as immutable objects, each with a unique identifier, metadata, and API access over HTTP. It is the model behind Amazon S3 and similar cloud storage services.
RAID
RAID (Redundant Array of Independent Disks) combines multiple physical disk drives into a single logical unit to improve performance, fault tolerance, or both, using techniques like striping, mirroring, and parity.
RAID 0
A data storage method that splits data across two or more disks in stripes to improve read and write performance, but provides no fault tolerance.