The first hurdle to understanding flash is being aware of the fact that there is no single form of flash; instead there are many.
When we talk about flash disks or solid state drives (SSDs), we are not talking about one type of memory but different kinds.
Here's a brief tour of some of the most important types of flash.
Hard disk drive form factor flash SSD
SSDs that use a form factor similar to a conventional hard disk drive are the most common on the market. They are deployed in large-scale enterprises and small businesses alike.
An SSD can do much the same job as a conventional hard disk drive (HDD) but instead of the data being stored in the grooves of a disk, it is stored on the flash memory chips. They work in a different way to the HDD because they will remember what is stored even after the power is switched off.
These flash memory chips differ from the flash memory in USB 'slot-in' drives in the type and speed of the memory. They are designed to be much faster and more reliable than the standard thumb drives. For that reason, they are much more expensive.
While an SSD may be installed in a server, it does not necessarily need to be of an enterprise-class drive. A good example is the boot drive used to start up a particular system.
Beyond boot drives, the major deployments for enterprise products are in a storage area network (SAN), network attached storage (NAS), and direct attached storage (DAS) systems.
"The drive format products have a great advantage of being able to plug-and-play into an existing array," according to the analysts Storage Strategies Now (SSN). "This does come at a cost over other implementations as drive format products need to pretend that they are really a disk drive."
This means that an industry-standard interface, such as Fibre Channel, Serial ATA, or Serial Attached SCSI, must be built into each drive so it can plug-and-play into existing products, such as drive bays or SANs.
PCIe format flash SSD
This format uses the full bandwidth of the CPU bus but, according to the SSN analysts, has some interesting tradeoffs. "Once the exclusive domain of Fusion-io, PCIe products are supplied by a large number of vendors, including Micron, Samsung, SanDisk and Toshiba," they wrote.
Last year in a prime example of market consolidation, Fusion-io -- whose original backers included Michael Dell and which boasted Apple co-founder Steve Wozniak as chief scientist -- was bought by SanDisk.
These devices offer very high performance but at a premium price point so they are only likely to be used by companies which have a particular need for the equipment and can justify the cost.
Another downside, according to SSN, is a lot of concern, "about the expense of redundancy and the potential for single points of failure taking very costly components offline [which] are associated with these devices".
All flash arrays
These storage arrays are purpose-built devices that are usually rack-mounted and act as either a shared block device or as a NAS file system. They can be built-up from available HDD format SSDs, DRAM memories, custom Flash modules, or combinations of some or all of these.
Appliance interface options for physical connections typically include Fibre Channel, InfiniBand, and Ethernet. "For physical connections SAN block-oriented connections, Fibre Channel Protocol over Fibre Channel, SCSI Remote Protocol (SRP) over InfiniBand, iSCSI over Ethernet and recently ATA over Ethernet (AoE) are popular options," said SSN.
NAS options typically include Network File System (NFS), Common Internet File System (CIFS), or Server Message Block (SMB) protocols running over Ethernet. Devices range in size and capacity from a single rack unit (Or 1U) to an entire rack supporting hundreds of TBs of solid state memory and there is plenty of choice.
They offer many advantages, especially their shareable nature, wide bandwidth, and scalability. The disadvantages include high centralised costs (all the eggs in one basket) and, unless they are carefully designed, they tend to have a single points of failure.
Hybrid HDD and flash array
Hybrid arrays that incorporate both flash devices and conventional HDDs are becoming increasingly popular. They offer lower cost per capacity by using the HDD for data that is accessed less frequently and using the flash components for data that is in more demand.
How do you work out which data to have in HDD and which in flash? Well, hybrid arrays usually incorporate advanced software algorithms for tier management, caching, data reduction, and data reliability. Also by using the higher capacity HDDs for mirroring and data striping, you can avoid data loss due to a single component failure.
Fabric connected PCIe flash
Using PCIe-based flash memories to accelerate server operations has become a regular practice within data centers of all sizes. Initially the use was no different from direct attached storage, except for being a lot quicker than the average hard disk drive.
This was OK as long as the data set being worked on could be kept within the capacity of the flash drive being used. If more capacity was needed, in order to keep any job within one unit of flash, a larger capacity flash drive would be needed.
So standard practice in the industry is now to keep 'in-demand' data on the flash card and use shared storage for everything else.
Help is at hand for the poor administrators in the latest generation of protocol, the non-volatile memory express (NVMe), which is used over PCIe.
This powers newer SSDs that can be available in multiple form factors, from standard 2.5-inch HDDs but can also be much smaller. The big breakthrough is that unlike the older protocols, NVMe supports hot-swappable devices.