Backups are Mission Critical
These days, it's hardly necessary to emphasize the importance of backing up your computer data; you don't have to ask too many people before someone will recount a lost-data horror story. Lightning strikes, fire, hacker attacks, viruses, software bugs, and operator error are some of the many possible causes of data loss. But the most common cause of catastrophic data loss is simple hard drive failure, which is easily overcome with a proper backup regimen.
Even in the best-case-scenario, losing data wastes the effort it took to generate it, which is a costly irritation; but if it causes a deadline to be missed, or creates logistical difficulties, then the consequences can be disastrous. Moreover, if the system used to generate and view the data is also lost, then additional time and expense must be spent recovering the system before one could hope to recreate the vital missing data.
For server-based networks, where many coworkers depend on one device, the consequences of data loss or system loss on the server are even worse. Thus it is imperative that the server is backed up on a routine basis, and that the backup medium is transported off-site to another building, and preferably stored in a fireproof box. Failure to implement such a system will eventually result in severe inconvenience, if not outright catastrophe.
Generally speaking, the frequency of backups determines the amount of data that could be lost; in other words: data recorded since the last backup may be unrecoverable in the event of some failure or disaster, and might require recreation from scratch. Therefore, it is wise to frequently backup vital data and system files, such as accounting data or operational databases.
In order to control costs, it is often appropriate to set up a rotating schedule of backups, in which we overwrite previous backups with a more modern copy of the data. However, this approach obviously prevents you from accessing that pesky file that you evidently deleted months ago. In order to achieve that capability, you could periodically take one of the backups out of the rotation (to be replaced by fresh storage media) and store it offsite as an archive. Alternatively, archives can be stored and backed up with the data, via scripting (this technique is particularly useful for accounting files).
Until now, we have been using the term 'data' in the general sense of all of the information on your network's various hard drives. But in more precise terms, your data are the files that your employees have directly created by using the system. It is the distilled electronic essence of your entire company's efforts, and it is the primary concern of any backup strategy.
If data backups are the first priority, then a close second is backing up the system configuration of critical devices. Data files on a backup tape are not useful in and of themselves; they must be restored to a properly configured system to be useful. If the hard disk in your server fails, you definitely want to have a recent backup of your data; but, ideally you would also want a complete backup of the operating system, applications, updates, configuration details, and data as well. In short, if you had an exact replica of the hard drive, then a disk failure would not necessitate a complete recreation of the system, and the significant downtime associated with such a recovery. That is the purpose of full system backups, they save labor and reduce downtime; consequently, it is wise to consider performing routine full system backups as a preventive maintenance.
This analysis begs the question: why bother with data-only backups when full-system backups can cover everything? To keep the answer short, full-system backups cannot be scripted, and they require more labor than data-only backups; so while it may only be financially prudent to perform full-system backups twice monthly, many of our customers prefer a more frequent schedule for data-only backups.
There is more to consider than just cost per gigabyte, when selecting a backup medium: What type of backup is it, data-only or full-system? How much data? How long does the backup need to be preserved? Is data security a priority? You should also consider logistical issues, such as how often will backups be performed, who will perform them, and whether the backup procedure necessitates system downtime. And the most important concern is how the data and/or system will be recovered if/when the backup is needed.
Actually, cost per gigabyte of storage media is probably your least important consideration. Various backup tape technologies cost anywhere between $0.50 and $2 per gigabyte; optical storage, like CD-R's and DVD-R cost around $0.50 per gigabyte; IDE hard drives cost around $1 per gigabyte. These prices will continue to drop over time, and the 'best deal' will vary from year to year, but there is no clear standout based on price. Magneto-optical and other exotic storage solutions are invariably more expensive per gigabyte; supposedly this premium is justified by their increased speed (we're not impressed); in any case they are uncommon and proprietary, and therefore not well-suited to the task of long-term data storage. The bottom line is to place convenience, compatability, security, logistics, and longevity ahead of any financial considerations, when contemplating your backup scheme.
So how do the various backup storage technologies compare? Here are some recommendations and guidelines:
For data-only backups, optical drives are the preferred method of archival storage. First of all, when properly cared for, they are extremely durable, so you can expect them to be viable for decades. Secondly, optical drives are ubiquitous, so you can recover your data to virtually any computer; moreover, you can be fairly certain that the disks that you burn now will be compatible with future technologies; avoiding obsolescence. Optical disks also present a data-backup method that is easy for the end-user to apply, which reduces labor costs. And finally, the storage size per disk is high enough to be quite useful, but low enough to keep individual disks cheap; this scenario is perfect for archiving data.
The venerable tape drive continues to be a factor in the backup market for one essential reason: tapes hold lots of data. We acknowledge that tape drives are the industry preferred full-system backup mechanism, but we do not see the wisdom of using a tape drive for anything other than data-only backups (and then only when you need to routinely archive huge amounts of data). Tape drives are expensive, slow, linear (not randomly accessible), proprietary beasts that need to be properly installed and configured before they'll do anything for you. If you're recovering a dead server from a tape, you have to reinstall the OS to get the tape drive to work, and then restore the rest of the data and system configuration; so what's the point of backing up the full system to a tape in the first place? And what are you going to do when the drive breaks and the manufacturer no longer supports it?
For all of these reasons, we generally steer our customers away from tape-based solutions. However, if you want to routinely archive lots of data (>12GB), or if you have an existing investment in a tape system, then we are happy to use the technology to its fullest. Tapes lend themselves well to scripting, which reduces the labor costs associated with tape backups; and tapes are very reliable archives on a 5-10 year basis.
Hard disk drives offer greater storage capacities than some of the most expensive tape drives, at commodity level pricing. Additionally, hard drives are randomly accessible and much faster than tape drives for even linear file access. And, most significantly, hard drives are bootable, so a full system backup to a hard-drive is an instant, fully functioning replacement if your server's hard drive is compromised.
In our opinion, hard drives are not suitable for archival backup (too fragile); and one should also be mindful of the labor costs involved with doing a full system backup. Nevertheless, backing up the full system to a hard disk is best medicine when you consider the potential recovery procedures. Note that mirroring the drives in a server (or utilizing other non-zero RAID levels) will backup the server, but only locally; this won't protect against fires, lightning, viruses, hackers, or an errant system update. We suggest taking a full-system backup drive offsite at least once a month, and supplementing that with more frequent data-only backups to optical disks (or, at least the network).