Solid-state storage options are evolving from consumer-level flash drives to highly capable flash-based storage arrays for enterprise data centers.
Flash storage is nonvolatile solid-state memory that presents to the server as traditional block- or file-based storage devices. The umbrella term "flash" encompasses devices ranging from a simple USB thumb drive to in-server solid state drives (SSDs) to an array-based enterprise storage system such as the IBM FlashSystem 810. Different devices use different interfaces and can vary in terms of write cycle rating and other features.
Enterprise users must match flash storage to suitable workloads. Flash storage uses about 20% of the power that disk drives need and reads about 100 times faster than disk storage. The advantages are clear for busy, read-centric workloads like databases. However, flash memory also has slow write speeds and limited write cycle life before becoming unreliable, so write-heavy applications will see latency increase.
Flash storage cells become unreliable after a finite number of write cycles or program/erase cycles. Flash memory devices are commonly rated to 100,000 write cycles, though newer flash chips hit 1 million write cycles. This resilience will continue to improve in enterprise flash devices. Several tactics have evolved to extend the life of flash storage systems. With the most popular, wear leveling, the storage device dynamically remaps storage blocks each time a write occurs. With the effect of write cycles spread across all of the memory, frequently written blocks won't fail prematurely. Many flash storage systems also rely on bad block management, wherein spare blocks are remapped in to replace failed blocks. Other memory integrity techniques include chip-level checksums and error correcting code.
Flash storage modules used for redundant array of independent disks (RAID) groups can deploy RAID 0 (striping) to aid in wear leveling or other RAID levels for more resiliency.
Stand-alone flash storage systems
Most data center flash resides within servers, but flash storage is increasingly deployed in separate enterprise storage systems. IBM offers the FlashSystem 710 and 810; Nimbus Data Systems makes the E-Class and S-Class flash storage systems; Tegile has Zebi storage arrays, and there are others.
Connectivity differentiates standalone enterprise storage systems. The Zebi HA2800 with 4.4 TB of usable flash storage capacity on a 2U rack provides 12 gigabit Ethernet ports for connectivity. The unit can be configured for dual-port 4/8 Gbps Fibre Channel or dual-port 10 gigabit Ethernet. It also supports standard storage protocols including storage-attached networks like iSCSI and Fibre Channel or network-attached storage like NFS and CIFS. The Nimbus S-Class flash memory system provides up to 100 TB of storage capacity with connectivity on up to 16 gigabit Ethernet ports, up to 16 ports of 10 gigabit Ethernet, up to eight ports of 8 gigabit Fibre Channel or up to eight ports of 40 gigabit quad data rate InfiniBand. For top flash storage system performance, you may have to upgrade the existing data center local-area or storage network.
Before settling on a flash-based enterprise storage system, verify that it supports the operating systems and hypervisors in your data center. For example, the Nimbus S-Class supports Windows Server 2003/2008, Linux, Solaris, HP/UX and AIX along with Hyper-V, ESX/ESXi, XenServer and Virtual Iron hypervisors.
Flash storage systems can support various data center environments, but wise data center administrators will investigate the storage product's requirements and specs, then ensure that the data center can accommodate it over the long term. Adopting a flash storage system intended for one OS and hypervisor is problematic if the business later shifts to another OS, hypervisor or storage network architecture.
Proof-of-principle projects will help you gather workload performance benchmarks for comparison against the same workloads using traditional storage.