This page is about theory, the theory behind SAN standards and SNIA / CIM models. If you are not a theory person, skip this page and go to the practical stuff in the rest of the SAN section.
SAN Evolution
SANs became a practical proposal in 1999, and were mainly used to connect a tape silo to several servers for backup. Switched Fabric SANs with auto failover followed and were used for storage consolidation as it was then possible to connect several servers to one, or a group of storage devices. Network Attached Storage (NAS) is also used for the same function.
However, NAS and SAN did not solve all of the problems of distributed data. One of the biggest issues is that each storage vendor tended to make devices that did not work together, so they needed different management methods and in the worst case, different SANs. This problem has been resolved to some extent by the introduction of standards; standards for storage devices, interfaces and management software. There are lots of standards bodies, which is a problem in itself, but the main ones are the Distributed Management Task Force (DTMF) and the Storage Networking Industry Association (SNIA). Some others include the Fibre Alliance, the SCSI Trade Association and the Fibre Channel Industry Association.
DTMF came up with the Common Information Model (CIM) that was basically about simplification of the management of distributed systems. Quoting from the DTMF website, "CIM's common definitions enable vendors to exchange semantically rich management information between systems throughout the network." Translated into English, this means that software and hardware products that are CIM compliant can talk to each other and understand each other.
SNIA took the CIM model as it applied to Storage Management and developed the Storage Management Initiative (SMI). In brief, SMI defines persistent naming standards and discovery systems so you can find a device; communication transports so you can talk to a device, and resource locking facilities so you can share a device.
New devices should be CIM/SMI compliant. Older devices can use CIM agents or Object Managers to make them compliant. A CIM agent takes a proprietary interface from a device and translates it into CIM format. A CIM Object Manager simply does this work for several devices.
The sections below show the relevant SNIA models for SAN virtualisation. SNIA of course has models for all the different aspects of storage.
In-band, Block Aggregation or Symmetrical Virtualization
Most storage vendors subscribe to the SNIA models for virtualization, and the SNIA model for symmetrical virtualization (SVZ) is shown below. Note that in this model the virtualization (or aggregation as SNIA call it) can happen in the storage device, in the host, or in the Storage Network. Most vendors seem to prefer to develop products that live in the latter. The reasons for this seem to be that if the intelligence is in the Storage Network then the physical storage can be isolated from the servers, and common utility services like remote mirroring and data replication can be developed that are independent of the underlying storage hardware. This means that virtualization in the SAN can permit mirroring or instant copy between disks from different vendors.
Out-of-band, File Aggregation or Asymmetrical Virtualization
Asymmetrical virtualization (AVZ), otherwise known as file aggregation or out-of-band virtualization, is when the virtualization appliance is not in the data path. The industry (or maybe just I) used to consider that AVZ and SVZ were mutually exclusive. This is changing as asymmetrical virtualization now considered as a means of file sharing and unique file identification across the SAN. As such, it complements SVZ and can be used with it.
File aggregation is similar technique to block aggregation. In the SNIA model below, hosts extract the file metadata from the file system to establish the path to the data, and then they access the data directly. This usually requires a metaserver to hold the metadata.
An example of AVZ is a SAN File System, a common file system specifically designed for storage networks. By managing file details (via the metadata controller) on the storage network instead of in individual servers, the SAN File System can move the file system intelligence into the storage network where it can be available to all application servers. File level virtualization aggregation provides immediate benefits: a single global namespace and a single point of management. This eliminates the need to manage files on a server-by-server basis. A global namespace is the ability to access any file from any client system using the same name.
XAM
SNIA is currently working in the next phase of SMI-S, called XAM, which should be finalised in 2006 with compliant products released in 2007.
XAM is about standardising ILM management. Storage Hardware APIs will be made partially vendor independent at application level, specifically for; fixed content storage, consistent metadata specification and handling. Basic search queries will be standardised, as will be management and security. SNIA specifications will mean that ILM metadata will be made vendor independent in terms of location and field availability, so ILM policies can be applied as standard over any compliant vendor's hardware or software. This will be especially important to avoid vendor locking for long term data retention.
The next page shows how these SNIA concepts work in practice.