The storage capacity of the humble tape cartridge has increased dramatically over recent years. In 1984, IBM introduced the 3480 Cartridge Subsystem, with a capacity per cartridge of 200MB and transfer rate of 3MB/s. Since then things have moved on a pace and the latest LTO-2 tape drives quote capacities of 200GB/s uncompressed and 40MB/s transfer rate. However, over the same period, disk capacities have also increased and price per gigabyte has dropped to almost £1 per GB for basic IDE drives. Now we have myriad disk replication technologies, is it time to put tape out to pasture?
New technologies have been developed to reduce the dependency on tape backups. For enterprise-class subsystems, there is synchronous or asynchronous mirroring. Data is written both locally and to a remote site in real time (or near enough to real-time for asynchronous links) and represents a disaster backup of the entire environment. Examples of this are RDF from EMC, Truecopy from HDS and PPRC from IBM. These technologies provide a complete disaster recovery solution where total or partial loss of a site occurs. Data at the remote site is guaranteed to be in-sync with the primary site, as a write request is not confirmed to the host until it has been successfully written both locally and remotely (obviously asynchronous writes do not guarantee to have written the remote I/O). This kind of synchronous mirroring has eliminated the need for large enterprises to use tape for their disaster recovery process. Instead, companies build another data centre and replicate data over fibre or ESCON links. The only disadvantage to this DR solution is the latency on the I/O caused by the remote write, which dictates the distance to the remote data centre.
To protect against local data issues, there are technologies such as Timefinder from EMC and ShadowImage from HDS. These provide a point-in-time copy of data (usually at a physical volume level), effectively creating a checkpointed copy of data, for instance a copy of a database or critical file system. Point-in-time backups can be used for fast data recovery from logical corruptions or accidental deletion of files. They can also be used for problem diagnosis or to replicate data when building new systems. All of this can be achieved much faster than using tape to perform the same function. For Networked Attached Storage, technologies such as Snapshots provide the equivalent functionality to point-in-time backups. Some products, such as the Netapp Filer take a snapshot by copying the pointers mapping out a file, creating an instant duplicate or backup of the data. Any amendments to the file are tracked by re-writing only the changed blocks, allowing point-in-time backups to be retained at the file level with minimum storage overhead. This allows instant file recovery without resorting to tape restores.
Whilst these technologies sound good for disk vendors, they appear to remove the dependency on tape - or do they? It is true to say that today it is impractical to consider using tape backups for full disaster recovery restores. However, tape has plenty features that still work in its favour.
Firstly, it is cheap. A single LTO cartridge capable of storing 100GB (uncompressed) can be picked up for about £50, lower in quantity. Excluding infrastructure charges, such as tape drives, that equates to 50p a gigabyte or less, depending on compression ratios. This makes tape half the price of the cheapest IDE disks currently on the market (as previously discussed at about £1 per GB) and much more cost effective than enterprise-class disk subsystems. Second, tape media is portable. Inactive tapes can be racked or stored offsite, significantly reducing the monthly management cost per GB. This also means that tape data doesn't need ongoing infrastructure to support it in the same way as disk subsystems do, in terms of power, cooling and subsystem infrastructure. Third, tape is a serial access medium, making it more difficult to delete data on a tape (although tapes have other security drawbacks - they can be stolen, lost or damaged) and making it easy to re-use once data on a tape has expired.
So is it almost the end of the line for tape? At the moment, probably not. For example, tape represents best option for long-term archiving of inactive data. Tape will continue to be used where cost of backup is the most significant price factor over cost of recovery. In addition, the virtualisation of tape could lead to the emergence of new developments new tape technology and will see tape in our environments for many years to come.