Storage Terminology

Common Storage Terms

Volume

A logical unit of storage provisioned on a storage array. Volumes are presented to hosts as block devices. Modern storage arrays typically support thin-provisioning and dynamic resizing.

LUN (Logical Unit Number)

An identifier used to address a specific storage volume within a target. In iSCSI, each volume presented to a host is assigned a LUN number. The term "LUN" is often used interchangeably with "volume" in storage discussions.

Multipath

A technique that uses multiple physical network paths between a host and storage array for redundancy and performance. If one path fails, I/O continues over remaining paths. See Multipath Concepts for detailed configuration.

ALUA (Asymmetric Logical Unit Access)

A SCSI standard that allows storage arrays to indicate preferred paths for accessing volumes. Paths are designated as "active optimized" (preferred) or "active non-optimized" (available but not preferred). Multipath software uses ALUA information to prioritize paths.

Path

A single network connection between a host interface and a storage portal. Multiple paths provide redundancy and can be used for load balancing.

Failover

The automatic process of switching from a failed path to an active path when a connection fails. Properly configured multipath enables seamless failover without I/O interruption.

APD (All Paths Down)

A condition where all paths to a storage device become unavailable simultaneously. Also known as "All Paths Dead" in some contexts.

iSCSI-Specific Terms

iSCSI (Internet Small Computer Systems Interface)

A protocol that enables SCSI commands to be sent over TCP/IP networks. iSCSI allows block-level storage access over standard Ethernet infrastructure.

Initiator

The host-side component that initiates iSCSI connections to storage targets. In Linux, the open-iscsi package provides the initiator functionality. Each initiator is identified by a unique IQN.

Target

The storage array component that receives iSCSI connections from initiators. A target can present multiple LUNs to connected initiators.

IQN (iSCSI Qualified Name)

A unique identifier for iSCSI initiators and targets. Format: iqn.YYYY-MM.reverse.domain:identifier

Examples:

• Initiator: iqn.1994-05.com.redhat:server01

• Target: iqn.2010-06.com.storagevendor:array.12345abc

Location on Linux hosts: /etc/iscsi/initiatorname.iscsi

Portal (iSCSI)

An IP address and TCP port combination that provides access to an iSCSI target. Default iSCSI port is 3260.

Example: 10.100.1.10:3260

A storage array typically has multiple portals across controllers for redundancy and performance.

Session

An active connection between an iSCSI initiator and target. A session can contain multiple TCP connections for performance.

Discovery

The process of querying a target to learn which volumes (LUNs) are available. Discovery can use the SendTargets method or iSNS (Internet Storage Name Service).

CHAP (Challenge-Handshake Authentication Protocol)

An authentication mechanism for iSCSI that verifies initiator identity using a shared secret. CHAP can be unidirectional (target authenticates initiator) or bidirectional/mutual (both authenticate each other).

Interface Binding

The practice of associating iSCSI sessions with specific network interfaces. Required for proper multipath operation to ensure each session uses a designated NIC.

NVMe-TCP-Specific Terms

NVMe (Non-Volatile Memory Express)

A high-performance storage protocol designed for flash storage. NVMe reduces latency and increases IOPS compared to SCSI-based protocols.

NVMe-TCP (NVMe over TCP)

NVMe protocol transported over TCP/IP networks. Combines NVMe performance benefits with standard Ethernet infrastructure. Uses port 4420 for data connections.

NVMe-oF (NVMe over Fabrics)

The specification that extends NVMe across network fabrics. NVMe-TCP is one transport option; others include NVMe-RDMA and NVMe-FC.

NQN (NVMe Qualified Name)

A unique identifier for NVMe hosts and subsystems. Similar in purpose to iSCSI IQN.

Format: nqn.YYYY-MM.reverse.domain:identifier or nqn.2014-08.org.nvmexpress:uuid:<UUID>

Examples:

• Host NQN: nqn.2014-08.org.nvmexpress:uuid:12345678-1234-1234-1234-123456789abc

• Subsystem NQN: nqn.2010-06.com.storagevendor:array.12345abc

Location on Linux hosts: /etc/nvme/hostnqn

Subsystem

An NVMe storage entity that contains one or more namespaces. Analogous to an iSCSI target. A subsystem is identified by its NQN.

Namespace

An NVMe storage volume. Analogous to an iSCSI LUN. Each namespace within a subsystem is assigned a Namespace ID (NSID).

Controller

The NVMe component that manages connections and I/O between a host and subsystem. Each connection to a subsystem creates a controller instance.

Discovery Controller

A special NVMe controller (port 8009) that provides information about available subsystems. Used for automatic discovery of storage resources.

Portal (NVMe-TCP)

An IP address and TCP port combination that provides access to an NVMe subsystem. Default NVMe-TCP data port is 4420; discovery port is 8009.

Example: 10.100.1.10:4420

A storage array typically has multiple portals across controllers for redundancy and performance. Each portal can serve multiple subsystems.

Native NVMe Multipathing

Linux kernel-level multipathing for NVMe devices, enabled via nvme_core multipath=Y. Unlike dm-multipath (used for iSCSI), native NVMe multipathing presents a single device path (/dev/nvmeXnY) with the kernel managing multiple underlying paths automatically.

IO Policy

The algorithm used by native NVMe multipathing to select which path to use for I/O operations.

Options:

• queue-depth (Recommended): Routes I/O to path with lowest queue depth

• round-robin: Alternates between paths

• numa: Prefers paths on the same NUMA node as the CPU

Configuration: Set via udev rule or sysfs: /sys/class/nvme-subsystem/nvme-subsys*/iopolicy

Connection Parameters

ctrl-loss-tmo (Controller Loss Timeout)

NVMe-TCP parameter specifying how long (in seconds) to wait for a controller to reconnect before declaring it permanently lost. Recommended: 1800 (30 minutes) for production.

reconnect-delay

NVMe-TCP parameter specifying the delay (in seconds) between reconnection attempts after a connection failure. Recommended: 10 seconds.

fast_io_fail_tmo

iSCSI/multipath parameter specifying how quickly to mark a path as failed after an I/O error. Recommended: 5 seconds.

dev_loss_tmo

iSCSI/multipath parameter specifying how long to wait before removing a failed device entirely. Recommended: 30 seconds. Should be longer than fast_io_fail_tmo.

no_path_retry

Multipath parameter controlling behavior when all paths to a device are down. Recommended: 0 (fail immediately) to prevent I/O hangs.

Network Terms

MTU (Maximum Transmission Unit)

The largest packet size that can be transmitted on a network segment. Jumbo frames (MTU 9000) are recommended for storage networks to improve performance.

VLAN (Virtual Local Area Network)

A logical network partition that isolates traffic. Storage networks often use dedicated VLANs for security and performance isolation.

QoS (Quality of Service)

Network traffic prioritization. Storage VLANs should have high QoS priority to prevent I/O delays during network congestion.

Quick Reference Table