Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
Having now discussed the storage technologies used, we can examine how they are implemented. The most common storage configuration is simply that of having disks connected to a server. Each disk is effectively a stand-alone unit holding different data to any other connected disk. This configuration is often referred to as Just a Bunch of Disks (JBOD). Although it is a well-known approach, JBOD creates a single-point of failure. If the disk crashes, all of the data on the drive will be unavailable. There are few environments where such an occurrence is acceptable. Effectively managing storage is about more than providing space for users, it's about providing a responsive system that is also secure and reliable. JBOD systems provide few tools or options that allow response or reliability to be improved. With JBOD you can configure servers with faster and larger disks, but the underlying problems of throughput and the single point of failure remain.
To address the potential of a hard disk failure causing a loss of data, back in 1988, researchers at the University of California in Berkeley created a system called the Redundant Array of Independent Disks (RAID), as an alternative to the single drive approach which is referred to as a Single Large Expensive Disk (SLED). RAID combines multiple disk units into a single logical drive. This can increase performance by writing data onto several disks at the same time, though actual performance improvements depend on the RAID standard being used and the underlying storage technology. There are various RAID standards, referred to as levels, ranging from Level 0, which provides increased performance but no fault tolerance, to combination RAID levels which take the advantages of certain RAID levels and combine them with others to eliminate disadvantages. Here are the definitions of the RAID levels;
There are other RAID level combinations that are used, though these are the result of ingenuity and improvisation on the part of technicians rather than any ratified standards.
Although, depending on the RAID configuration, there are performance and fault tolerant advantages to using it, RAID only solves fault tolerance from a disk perspective rather than for the equipment it is attached to.
Whether JBOD or RAID, the next part of the storage puzzle is the way in which the storage subsystem is made available to clients. That can be in one of three ways - DAS, NAS or SAN.
Traditionally, all mass storage connected directly to the server, hence the term Direct Attached Storage (DAS). DAS configurations can be either a JBOD (Just a Bunch of Disks) or a Redundant Arrays of Independent Disks (RAID). The key aspect of DAS is that the disk subsystem is either in the same physical box as the system processor, or in an external box connected directly to the system processor unit. DAS satisfies some applications, but as access to the disks must occur through the server processor, each disk access takes valuable processing power away from the server. Further, each user must access the storage through a single channel that attaches the disk(s) to the server. This means that simultaneous disk accesses must be queued, and users can be left waiting. Enhanced caching systems alleviate some of these delays, but network performance lags when the data stores get large or when read and write requests take a long time to complete.