What Is iSCSI?

iSCSI (Internet Small Computer Systems Interface) is a storage protocol that wraps SCSI commands, the command set used by server drives, inside TCP/IP packets so they can travel over an ordinary Ethernet network. The Internet Engineering Task Force (IETF) published it as RFC 3720 in 2004, and the current specification is RFC 7143.

The computer that uses the storage runs an initiator. The storage system that holds the disks runs a target. The initiator logs in to the target over TCP, by default on port 3260, and is given access to one or more storage volumes, each identified by its LUN. From then on the operating system sees each LUN as a local disk: it partitions and formats it, installs applications on it and can even boot from it.

Because iSCSI works at block level, it builds a storage area network (SAN) rather than a file share. Each LUN normally belongs to one host at a time, unless a cluster file system coordinates access.

Setting up iSCSI means configuring both ends and the network between them. The main pieces are these.

iSCSI is one of several ways to build a block-storage network. Fiber Channel is the traditional SAN fabric. Fiber Channel over Ethernet (FCoE), defined by the INCITS T11 committee, carries Fiber Channel frames directly in Ethernet frames, and NVMe over TCP is the newest option.

FCoE needs Data Center Bridging (IEEE 802.1Qbb priority flow control and related standards) so that no storage frame is dropped, and it cannot cross a router. iSCSI relies on TCP to recover lost packets, so it runs on ordinary switches and across routed networks, at the cost of slightly higher latency and CPU load.

A NAS and an iSCSI target can be the same box on the same network, so the difference is easy to miss. A NAS shares folders: the NAS owns the file system, and many clients open files on it over SMB or NFS at the same time. An iSCSI target shares disks: each host receives a raw volume, formats it with its own file system and treats it as local.

That makes iSCSI the better fit for workloads that expect a local disk, such as databases, virtual machine stores, Windows applications that refuse network paths, and operating systems booted from the network. A NAS is simpler for shared documents, media and backups. Many storage appliances offer both, so a site can use iSCSI LUNs for its servers and a NAS share for its users from one system.

iSCSI performance depends on the network more than on the protocol. A few practices make the difference between a SAN that feels local and one that stalls under load.

Use a dedicated storage network or VLAN, at 10GbE or faster, and two independent paths with multipath I/O. Enable jumbo frames, usually a 9,000-byte MTU, end to end, on initiator, switches and target alike. Choose switches with enough buffer to absorb bursts: managed industrial Ethernet switches suit SANs in cabinets and plant rooms. Fiber uplinks avoid the distance limits of copper, and our FAQ on SFP+ vs RJ45 for 10GbE covers the trade-offs.

At the edge, iSCSI suits sites with several computers that share one storage pool, such as a small virtualization cluster in a plant room or depot. The compute nodes are often rack mount servers with 10GbE or faster ports, booting locally and keeping virtual machines on the shared LUNs, so a failed host can be restarted on another.

We build edge servers, switches and storage as one system as standard, with the network ports, iSCSI initiators and storage volumes configured before it ships.