Store wares: A new hope for data managers

Data storage vendors can rarely, if ever, afford to rest on their laurels when it comes to innovation. The pressure to squeeze more information into smaller components that offer higher capacity and faster read/write access while also using less electricity is unrelenting. Meanwhile, many IT and datacentre managers are frequently required to integrate legacy hardware and software into more efficient systems as they upgrade.

Disk technologies
The hard disks that populate the arrays inside storage area network (SAN), network attached storage (NAS) and servers need to deliver a constant balance between cost, capacity and performance, all of which vary according to the type of interface in use. Serial attached (SATA) is currently the lowest cost per gigabyte but disks are slower, whereas Fibre Channel (FC) and Serial attached SCSI (SAS) are faster but also more expensive.

Many organisations are now adding in solid state disk (SSD), which is considerably more expensive again, to existing RAID arrays to handle applications that have high I/O demands where transactions need to be completed particularly fast, as is the case with e-commerce web sites.

The latest generation of RAID arrays offer even greater flexibility in allowing different types of disks to be mixed together within the same rack. In this type of set-up, storage management, tiering and information life cycle management (ILM) software provide the automated tools that allow IT managers to store data on different types of disk to provide the right read/write speed, according to the application in use across SANs. Many hardware vendors, including Pillar Data Systems, NEXSAN and Compellent (acquired by Dell) seem content to mix Fibre Channel, SATA and SSD on the same array, for example.

Magnetic tape-based storage has been prematurely written off as an enterprise storage medium for at least a couple of decades, if not more, but the combination of legacy equipment and regular upgrades to backward compatible, larger capacity, faster media still convinces many IT departments that the technology is worth sticking with for now, if only for backup, archiving and disaster recovery rather than primary data storage purposes.

Ultrium linear tape open (LTO)-4 cartridges offering uncompressed capacities of 800GB are still being sold by HP, Tandberg and IBM, for example, with LTO-5 (1.5TB per cartridge) introduced last year and future versions through to LTO-8 on the cards.

Green storage
Tape is often touted as greener storage technology because it does not rely on constantly spinning disks, an argument also applicable to SSDs because they are based on a form of high-capacity Flash memory, which can be particularly important for datacentres and server farms that struggle to source all the power they require in the first place due to localised restrictions.

While storage technologies touting green credentials abound, nothing is more environmentally friendly than using less storage capacity in the first place. This is where compression and data de-duplication come in, though both can have negative effects on data access, backup and retrieval performance in some circumstances.

Data de-duplication eliminates multiple copies of the same data stored across the SAN, keeping a single master record or index, which is accessed on demand and which directs the application to where the original data is stored. It works best when used in conjunction with remote, disk-based backup, and resides either in a dedicated appliance or runs on a server. Many vendors, including EMC (which acquired Data Domain), Symantec, NetApp (Alacritus) and IBM (Diligent), are active in this market.

Depending on the type of file being processed, however, de-duplication compression rates can vary from one per cent to 50 per cent, according to reports.

Thin provisioning
Thin provisioning technology is designed to optimise storage utilisation, and is touted as a cost-effective alternative to the widespread practice among storage managers of over-provisioning.

Applications and systems sometimes crash if they run out of allocated storage space, leading storage managers to oversubscribe the volume of gigabytes or terabytes available to them to avoid performance problems and compensate for data growth. This results in a percentage of available disk space often lying idle and waiting for an application or user to fill it, a problem that thin provisioning is designed to avoid by pooling available disk space on multiple storage devices into a central reservoir, and releasing capacity only when it is actually needed or when utilisation reaches a pre-configured threshold.

Vendors offering both software- and hardware-based thin provisioning include Datacore, EMC and Hitachi Data Systems (HDS), but the technology has its limitations: deleted data must be reclaimed, certain file systems struggle to reuse released blocks of data, and some applications are configured to expect contiguous data or spread metadata across an entire volume, all of which can affect or negate the advantages of using the technology.

LAN and SAN converge
Network consolidation requirements have also prompted many IT managers, particularly those looking after large datacentres, to evaluate new network and storage “fabric” architectures from industry giants such as Cisco, Juniper and Brocade, which converge local (LAN) and storage area networks (SANs) into a single network, ostensibly optimised to support cloud-based services.

The attraction comes from being able to route data traffic from FC arrays and disks over IP-based Ethernet networks, using converged network adapters (CNAs) and fabric adapters (FAs). These offer simultaneous connectivity to a range of LAN, NAS and SAN equipment at speeds of up to 10Gbit/s initially, and they will eventually reach 100Gbit/s, using the Fibre Channel over Ethernet (FCoE) protocol. CNAs and FAs are intended to replace the network interface cards (NICs) and host bus adapters (HBAs) that were previously used for the same purpose on two separate networks within the datacentre.

The concept is not new, and it follows the same thinking as iSCSI, which was also designed to make FC within the SAN redundant. Tussles over standards and the high cost of components in the short term may yet mean that FCoE LAN/SAN convergence has a similarly limited impact.