IBM SAN Volume Controller

IBM SAN Volume Controller
Company type	Product
Industry	Data storage
Parent	IBM

The IBM SAN Volume Controller (SVC) functions as a block-level storage virtualization device within the IBM System Storage portfolio. It establishes a virtualization layer inside a Fibre Channel storage area network (SAN).

Architecture

The IBM 2145 SAN Volume Controller (SVC) operates as an inline virtualization appliance situated between servers and storage arrays. It appears to hosts as the storage provider, while simultaneously presenting itself to storage arrays as a host, connecting through one or multiple SAN fabrics.

This virtualization layer facilitates seamless, non-disruptive replacement of any component within the storage infrastructure, including the SVC units themselves, and streamlines compatibility across diverse server and storage platforms. Key advanced features are embedded in this virtualization tier, enabling effortless migration between different storage array vendors. Implementing SVC in a multi-site configuration (known as stretched clustering) delivers disaster recovery capabilities and ensures continuous service availability.

SVC nodes cluster together in groups ranging from 2 to 8 units, providing scalable performance proportional to cluster size. These rack-mounted devices, built on IBM System x hardware, incorporate redundant power supplies and built-in battery backups. Each node is equipped with Fibre Channel ports dedicated to inbound, outbound, and intracluster communications; additionally, hosts can connect via FCoE and iSCSI Ethernet interfaces. Intracluster links preserve cache coherence, exchange status information, and route data; these connections require zoning for security.

Write cache data is mirrored between paired nodes known as I/O groups. Virtualized volumes are distributed among I/O groups to optimize performance and can be migrated without interruption during maintenance or technology upgrades. Both nodes in an active pair handle write operations concurrently; other cluster members forward read/write commands to the corresponding I/O group. In the event of failures, automatic failover occurs seamlessly within or between I/O groups. Hosts must utilize multipath drivers such as IBM Subsystem Device Driver (SDD) or standard MPIO implementations.

The SVC platform is based on the COMmodity PArts Storage System (Compass) architecture developed at IBM's Almaden Research Center, with most software components created at IBM Hursley Labs in the United Kingdom.

Terminology

Node – an individual 1U or 2U server unit.

SVC node models
Model-Type	Cache Capacity [GB]	Fibre Channel Speed [Gb/s]	iSCSI Speed [Gb/s]	Based On	Release Date
2145-4F2	4	2	n/a	x335	2 June 2003
2145-8F2	8	2	1	x336	25 October 2005
2145-8F4	8	4	1	x336	23 May 2006
2145-8G4	8	4	1	x3550	22 May 2007
2145-8A4	8	4	1	x3250M2	28 October 2008
2145-CF8	24	8	1	x3550M2	20 October 2009
2145-CG8	24	8	1 (optional 10 Gbit/s)	x3550M3	9 May 2011
2145-DH8	32	8 & 16	1 (optional 10 Gbit/s)	x3650M4	6 May 2014
2145-SV1	64...256	16	10 Gbit/s	Xeon E5 v4	23 August 2016
2147-SV1	64...256	16	10 Gbit/s	Xeon E5 v4	23 August 2016

I/O group – a duo of nodes that replicate each other's write operations.
Cluster – an assembly of up to four I/O groups functioning as a unified system.
- Stretched cluster – a disaster recovery configuration where I/O groups span two separate locations plus an additional witness site.
- Cluster IP address – a single IP used for cluster management via (secure shell and HTTPS).
- Service IP address – unique IP allocated per node for maintenance purposes.
- Configuration node – stores cluster settings and manages the cluster IP address.
Master Console (SSPC) – graphical interface for administering SVC prior to version 5.1, running on a dedicated workstation.
- From release 6.1 onward, management is performed through a web-based HTML5 GUI hosted directly on the configuration node.
Virtual Disk (VDisk) – logical storage presented to hosts; referred to as Volume in version 6's interface.
Managed Disk (MDisk) – physical LUNs from external disk arrays that are virtualized by SVC; form the basis for image mode VDisks.
Managed Disk Group (MDisk Group) – collection of MDisks whose extents compose striped or sequential mode VDisks; termed Pool in version 6's GUI.
Extent – smallest allocatable storage segment; MDisks are subdivided into extents which collectively constitute VDisks.

Timeline

SVC models were made available shortly after their announcement. The light green bars represent the order windows, while the light blue bars indicate the standard service period following withdrawal. Information is accurate as of August 2019. Although service terms vary between models 2145 and 2147, their hardware remains identical.

Performance

The SVC Release 4.3 established a new world record in the Storage Performance Council (SPC) SPC-1 benchmark, achieving nearly 275,000 (274,997.58) IOPS, making it the fastest storage subsystem at that time (October 2008). Additionally, the SPC-2 benchmark demonstrated throughput exceeding 7 GB/s.

With Release 5.1 in March 2010, previous records were surpassed: a 4-node cluster delivered 315,043.59 SPC-1 IOPS, while a 6-node cluster reached 380,489.30 IOPS. These records remained unbroken until October 2011.

Release 6.2 set a fresh SPC-1 world record in January 2012 by reaching over 500,000 (520,043.99) IOPS using eight SVC nodes paired with Storwize V7000 backend storage.

The enhancements introduced with Release 7.x included support for additional CPUs, expanded cache capacity, and more adapters. Its optimized cache operates with a fall-through latency of 100μs and a cache-hit latency of 60μs, allowing the SVC to front IBM FlashSystem solid-state storage without noticeable performance degradation.

Included Features (7.x)

Virtualization and Mapping: Servers interact with the SVC as though it were a dedicated storage controller; virtual LUNs correspond to volumes carved from pooled managed disks (MDisks). Virtual capacity can surpass physical limits, reaching up to 32 PB depending on extent size.

Data Migration and Pooling: Volumes can be moved between capacity pools (MDisk groups) while maintaining active I/O and caching, facilitating smooth hardware migration or upgrades without requiring data duplication.

Image Mode Import/Export: Enables seamless non-virtualized pass-through of existing client data LUNs into or out of the SVC cluster.

Fast-write Cache: Host write operations are acknowledged once data is committed to mirrored cache prior to destaging to storage controllers, boosting performance particularly in midrange setups.

Auto Tiering (Easy Tier): Automatically relocates frequently accessed "hot" data to SSDs and less active "cold" data to spinning disks based on usage patterns. This optimizes workloads even without SSD presence and balances performance across Enterprise- and Read Intensive Flash media.

Solid State Drive (SSD) Support: Supports both external SSD devices and internal slots (up to 32 per cluster), enhancing spinning disk pool performance through Easy Tiering.

Thin Provisioning: Allocates physical storage only when data is written, discarding zero blocks during import or migration to optimize space utilization. Works in conjunction with FlashCopy for efficient snapshot space management.

Virtual Disk Mirroring: Keeps redundant LUN copies across different storage controllers to safeguard data integrity.

Site Protection with Stretched Cluster: Allows geographically dispersed clusters spanning up to four datacenters within a 300 km radius, presenting a single logical instance with synchronous read/write access from both sites without failover or role switching. Integrates with Live Partition Mobility and VMotion for virtual server migration without bulk data movement.

Geographical Crossover Access: Every node within a Stretched Cluster has remote read/write access to mirrored storage, minimizing site resynchronization times following failures.

Hot Standby Nodes: Nodes powered on and prepared to rapidly replace failed nodes within stretched or local clusters.

Enhanced Stretched Cluster: Improves data path efficiency in metro or geo-distance clusters by reducing cross-site traffic and prioritizing shortest paths for read/write operations.

Stretched Cluster with Golden Copy (3-site DR): Maintains an additional synchronous or asynchronous copy on separate clusters for disaster recovery against metro-area failures using licensed Metro- or Global Mirror features.

Hyperswap: Ensures uninterrupted failover of data access between geographically separated IO groups or clusters with simultaneous write acceptance on both ends. Compatible with native multipath drivers supporting ALUA and integrates with Live Partition Mobility or VMotion. Requires Metro Mirror license and at least two pairs of nodes.

Transparent Cloud Tiering: Utilizes Swift- and S3-compatible object stores as cold tiers for incremental snapshots and archival without impacting live production workloads. Supports both on-premises OpenStack Swift and off-premises Amazon S3 or Softlayer with AES encryption as a licensed feature. Provides cost-efficient archiving comparable to tape media pricing.

Optional Capabilities

A range of optional features are available under separate licenses, usually on a per-terabyte basis:

Real-Time Compression: This technology reduces data size during transmission, delivering at least a 50% reduction in storage footprint and reaching up to 80% compression specifically for Oracle databases. It leverages specialized compression hardware to avoid impacting system performance, making it ideal for intensive database environments. Due to the algorithm’s use of temporal locality, read speeds can improve on certain data types, such as SQL databases stored on spinning disks. The compression efficiency is comparable to "zip" (Lempel–Ziv–Welch) algorithms with a large dictionary and can be reliably estimated over petabyte-scale datasets using the Comprestimator utility. Real-Time Compression is compatible with Easy Tiering, Thin Provisioning, and Virtual Disk Mirroring. This feature originated from Storwize Inc., an acquired startup whose name also inspired the IBM storage systems family derived from SVC.

FlashCopy (Snapshot): Allows for creating point-in-time disk snapshots useful for backups, rollbacks, or testing individual volumes. Snapshots consume only the changed data space unless the target volumes are fully provisioned. FlashCopy supports three modes—Snapshot, Backup volume, and Clone—the latter automatically detaches from its source volume. It relies on optimized copy-on-write mechanisms rather than journaling and uses a bitmap with adjustable grain size. A single source volume can have up to 256 concurrent targets that may be incremental or arranged in cascaded tree structures. These targets can be reapplied either to their original source or to different volumes, including those with resized capacities.

FlashCopy rollback (time machine): Offers rollback functionality with fine-grained consistency points across multiple LUNs. This feature requires both a FlashCopy license and Spectrum Control Snapshot software.

Metro Mirror – synchronous remote replication: Facilitates disaster recovery by replicating data synchronously to sites located within approximately 300 kilometers.

Global Mirror – asynchronous remote replication: Supports disaster recovery across distances spanning thousands of kilometers. It can be configured for links with high latency and either low or high bandwidth while maintaining consistent recovery point objectives (RPO) under one second.

Global Mirror over IP – remote replication via Internet: Utilizes SANslide technology embedded in SVC firmware to optimize data mirroring traffic over TCP/IP networks, potentially increasing transfer speeds by up to 100 times over extended distances.

Encryption of data at rest: SVC and other Spectrum Virtualize platforms provide transparent encryption of data stored locally, virtualized across storage tiers, or in cloud environments by default using 256-bit AES-XTS encryption. Encryption keys can be generated and stored locally on removable media or managed through key lifecycle services—these two options are mutually exclusive.

Other Products Utilizing SVC Software

On October 7, 2010, IBM launched the IBM Storwize V7000, marking the debut of the Storwize product line. This system combines SAN Volume Controller (SVC) software with built-in storage hardware to offer a mid-tier storage solution. Smaller variants such as the Storwize V5000, V3700, and V3500 provide streamlined configurations with reduced cache sizes, processors, adapters, and fewer features.

The IBM FlashSystem V9000 integrates SVC firmware with IBM FlashSystem solid-state storage enclosures.

In 2015, IBM rebranded its virtualization software as Spectrum Virtualize to align with its software-defined storage vision and highlight enhanced interoperability across platforms.

SVC code utilized in non-IBM products

The Actifio Protection and Availability Storage (PAS) appliance integrates components derived from SVC software to facilitate extensive compatibility. This PAS system delivers capabilities for backup, disaster recovery, and maintaining business continuity.

External links