Software has become the new battle ground of storage. The major vendors – EMC, Hewlett-Packard, Hitachi Data Systems, IBM and others – all acknowledged that the key differentiation, the true competitive edge will in future come not from faster disks or denser platters, but from tools for managing the vast quantities of data stored across the line up of disk systems, tape libraries, optical jukeboxes and the networks that connect them.
That has created two distinct markets within one. One is dominated by the search for efficiency, and is largely centred on efforts to connect storage devices in networks. The other is in a more traditional area – backup and recovery – and one where new pressures are demanding anything but traditional answers.
For most vendors, though, storage area networks (SANs) – and their main themes of virtualisation and storage resource management (SRM) – are where the primary action is. By allowing multiple servers to share different storage devices, SANs provide a pool of shared resources that can be allocated to applications as needed.
SRM aims to automate much of the labour-intensive and error-prone processes associated with allocating storage resources, while virtualisation enables the servers and their applications to treat the devices on the storage network as a single pool of capacity. In theory, virtualisation should enable applications to address any storage device in the network, no matter which manufacturer it comes from, but it is only in the past few months that standards have matured to a point where vendors have started to make commitments to support such interoperability in their core products.
By allowing more efficient and flexible use of storage resources with less manual involvement, virtualisation and SRM promise significant improvements in IT staff productivity, as well as much better utilisation of the storage resources that might have previously been tied to a single server.
And the attraction of such approaches is showing up in demand analyses. During the past two years, SRM and virtualisation software have become major parts of the storage management software spectrum. According to research by Dataquest, organisations will spend $1 billion on SRM products this year, four times more than in 2002. Similarly, they will spend $728 million on virtualisation software software, up from a mere $156 million in 2002. And taken together over that period, these two technologies’ share of the overall storage management software market will go from a 8% to 17%, says Dataquest.
Efficient management of storage resources counts for nothing if a specific problem makes that stored data unavailable. The other half of the storage management software story is all about the mechanisms for maximising availability. Principally that means backup and recovery, the bedrock of storage management software, but it also extends to related areas such as archiving and data replication (and its many guises of copying, mirroring, shadowing and snapshotting data).
Backup has traditionally been done by connecting a secondary unit directly to the server or to the local network the server is attached to. But with growing data volumes, these approaches have become increasingly problematic, as backup traffic occupies an ever larger proportion of bandwidth and the window for copying narrows. In fact, in many organisations, especially those transacting over the web, the window is now almost negligible.
By its very nature, a SAN overcomes these problems; backups are carried out on a separate network from that used by the application servers, and storage capacity can be sourced from elsewhere on the network while the backup is underway.
For backup to be of any use, of course, organisations need to be able to quickly recover their data when necessary. Although tape is today the main mechanism for backup and retrieval, it is regarded as too slow for restoring the mission-critical data of many organisation’s applications.
Given that, there is a more immediate form of backup allowing virtually instant recovery: mirroring.
With mirroring, all data generated by an application is written to disk twice simultaneously. So, in theory, any lack of availability can be obviated by the fact that at any one time the same set of data can be accessed from two different disk systems.
Mirroring within the same building guards against failure of either system, but not against any catastrophic local event that knocks both out. Remote mirroring on the other hand provides an additional level of insurance by ensuring that the two copies are maintained some distance apart.
The duplication involved in mirroring does make it expensive, but for certain applications (bank transaction, airline reservations and the like), the cost can be justified.
Remote copying is a less expensive alternative. Here a copy is taken at intervals, rather than constantly, and recorded at a remote location. Should the first site go down, processing can restart at the second site from the point the last copy was taken.
Yet another technique is to take point-in-time copies at periodic intervals. This is done by taking a ‘snapshot’ of the data store’s contents. But by only tracking the changed elements in the database rather than copying the whole thing, snapshots are considerably less expensive – though they do require careful version control.
Until the advent of SANs, software for replication products was all supplied with the disk system, and was inherently proprietary. As a result, any backup had to be made on the same type of disk system used for holding the original databases. Now it is possible to place such backup software in the storage network, and get it to work cooperatively with virtualisation software.
That gives rise to the situation where the two independent strands of storage management – SAN-enabling software and backup – are now overlapping. And even though the $10.4 billion storage software market has doubled in size in the past two years, the two will be the major drivers as spending on storage software rises to $16 billion annually during the next three years.