Section 9.1. Decision Factors

9.1. Decision Factors

As mentioned previously, there are only eight decision factors to consider when deciding which backup drive to buy: reliability, duty cycle, transfer speed, flexibility, time-to-data, capacity, removability, and cost. You need to look at each of the eight decision factors and decide which are most important to you and your data. For example, an environment with a 6 TB database might care about reliability first, transfer speed second, and cost last. On the other hand, an environment that is going to use some type of Hierarchical Storage Management (HSM) system will be concerned about reliability first, time-to-data second, and transfer speed last. Let's take a look at these eight decision factors now.

9.1.1. Reliability

Any electronic repair shop will tell you that moving parts fail. If a mechanism to be repaired contains 25 circuit boards and 1 spinning wheel, they always check the spinning wheel first. The bad news is that backup drivesand especially tape driveshave more moving parts than any other part of your computer system. This means that, statistically speaking, a backup drive has a much greater chance of mechanical failure than a CPU.

Cheap tapes can cause data loss. Don't buy cheap tapes!

Drives that fail on a regular basis affect your system availability because replacing some drives requires shutting down the system. Unfortunately, what usually happens is that the replacement of malfunctioning drives is very low on the priority list. A bad drive may wait several days or even weeks to be replaced, even if the replacement part is available immediately. (This is because swapping drives sometimes requires system down time.) So, if drives fail too often, their failure can significantly affect the overall integrity of the backup system. That is, drives could fail so often that a large backup or restore would not have enough drives available to complete within a reasonable timeframe. This is why drive reliability is a very important factor in deciding what backup drive to get.

Drive manufacturers use the mean-time-between-failure (MTBF) value to represent the reliability of their drives. Unfortunately, these numbers usually are derived from artificial environments that attempt to simulate thousands of hours of use. (How else would a drive that has been on the market for only six months be able to advertise a 30,000-hour MTBF? That's almost 3.5 years.) The best source of information about the reliability of different drives is the Internet. Use several search engines to locate discussions about the backup drive in question. Google is definitely helpful here. Search its complete archive for different phrases that might locate discussions about the drive you are considering (e.g., DLT, LTO-4, T10000, and so on). If you don't find any, post your own message and see who replies. (The most useful Usenet group for this topic is comp.arch.storage. You do remember Usenet, right?) After that, you should do the unthinkableactually talk to people. Before you buy a new type of backup drive, I would talk to no less than five companies using that drive.

Desperate Times

I have been saved a number of times with the "twist-start" recovery method. You know, when you get a hard drive that doesn't seem to be spinning up right, and you give the drive a quick twist of the wrist. Magically it gets up to speed and works for a little while.

Brian O'Neill

While the removability of the media is an important feature of tape drives, it requires the drives to be an open system. Dust and other contaminants are introduced in the drive every time the door is opened to insert or eject a tape, and contaminants are the death of any mechanical system. In contrast, disk drives are closed systems; the media is never separated from the drive, and no air is ever allowed to enter the unit. This is a major reason why disk drives are inherently more reliable than tape drives, which is why they are becoming increasingly popular as backup drives.

Finally, another reason why disk drives are more reliable than tape media is that you can use RAID to mitigate the risk of a single piece of media causing you real damage. With tape or optical, you make a backup, set it on the shelf, and hope the bits don't fall off or get rearranged while it's sitting there. You never really know, it's any good until you use it for a restore. With RAID-protected disk, you can constantly monitor the health of a piece of media. If any disk drives start to look unhealthy (or fail altogether), just replace them, and have the RAID system rebuild them. No lost data, no failed backupsjust a little maintenance.

9.1.2. Duty Cycle

When comparing the MTBF of different drives, don't forget to look at the duty cycle the drive was designed for. A drive's duty cycle measures the percentage of time a backup drive spends reading, writing, and verifying data. For example, if a particular drive operates for 6 hours a day, that drive has a 25 percent duty cycle. If it operated continuously, it has a 100 percent duty cycle.

Each drive has an intended duty cycle. Fibre Channel and SCSI disk drives, for example, are usually designed for applications demanding a 100 percent duty cycle; ATA disk drives are designed for lower duty cycles. This is why Fibre Channel and SCSI disk drives are often used in high-volume applications, such as database servers or very busy file servers. ATA disk drives have historically been used in PCs and laptops where the demands on the disk drives are much lower. Even now that they're being used in large storage arrays and virtual tape libraries, these storage systems are usually intended for reference data. Backup, archive, and imaging systems create reference data. You make backups every day, but any given backup is rarely accessedif at all. The same is true of imaging and archive systems. If your company creates an image of every customer contract, for example, any given contract is stored once and only occasionally (if ever) accessed.

Tape drives also have intended duty cycles. Mainframe-style tape drives are typically designed with a 100 percent duty cycle whereas those intended for open-systems markets are designed for lower duty cycles. This is because mainframes typically use their tape subsystems as a secondary storage system while open systems typically use tape drives only for backup. Mainframes are constantly reading and writing production data to and from tape. Therefore, their tape drives must be prepared to operate continuously, and mainframe tape libraries are judged on how many exchanges they can perform per hour. In contrast, most open-systems backup applications use their tape drives for less than half the day, which is why tape drives designed for open systems typically have much lower duty cycles.

It's important to buy a tape drive with a duty cycle that fits your application. Not doing so can significantly increase the cost or decrease the reliability of your backup drive. If you buy a drive with a 40 percent duty cycle and use it continuously, just expect that drive to fail much sooner than its published MTBF. If you have an application with a 40 percent duty cycle, and you buy a drive that was intended for a 100 percent duty cycle, expect to pay as much as 300 percent more for the same speed and capacitysometimes even less speed and capacity! The vendor with the drive that has a 100 percent duty cycle may claim their drive is more reliable, and they're right. If that drive was used in a 100 percent duty cycle environment, it will be more reliable than the drive with a 40 percent duty cycle used in that same environment. However, a tape drive designed for a 40 percent duty cycle should be just as reliable as the 100 percent duty cycle drive if it is used in a 40 percent duty cycle environment, such as traditional backup.

9.1.3. Transfer Speed

Transfer speed is also very important to consider when deciding on a backup drive. How fast can the drive read and write data? When comparing different drives, of course, you must compare native transfer speeds. A drive's native transfer speed is its speed without compression. This often is difficult to assess, because many drives quote only the compressed numbers. They usually attach a footnote to that number that says something like, "This number assumes a 2-to-1 compression ratio. The actual transfer speed varies based on the compressibility of the data." The reason for this is that different manufacturers make different claims regarding compression ratios. Some vendors have claimed a typical compression ratio of as much as 5 to 1. While it is true that different compression algorithms do yield different results, it is very difficult to verify a particular vendor's claims of better compression. To make sure that you are comparing apples to apples, always compare native transfer speeds.

Some vendors do not use the term native transfer rate. They might use the term head-to-tape transfer rate, which refers to how fast the recording head can write the data on the tape. This rate does not change with compression. If the data is compressed prior to being sent to the recording head, the drive's effective throughput is increased, but the head-to-tape speed does not change.

When looking at drive specifications, try to locate the word sustained. Sustained transfer rates provide the fairest comparison between drives. Some drives quote burst rates and synchronous rates, which are both temporary, best-case scenarios. (Based on your application, you may wish to compare burst and synchronous transfer rates as well; just make sure you know that's what you're doing.)

Don't forget to put everything in the same terms. Some drives publish their numbers in gigabytes or megabytes per hour. To get megabytes per second from megabytes per hour, just divide the number by 3,600, which is the number of seconds in an hour. If it's gigabytes per hour, you can get megabytes per second by dividing the number by 3,686,400, which is 3,600 times 1,024.

9.1.4. Flexibility

A backup drive can be flexible in two different ways: it can respond well to different data rates, and it can be used in different ways. Generally what you find is that tape and optical drives are not very flexible, and that disk drives are fairly flexible. Let's explore this idea.

9.1.4.1. Tape drives: Not so flexible

Tape drives do not generally respond well to different data rates, they can be used for only one purpose, and you generally need at least one per backup server.

Tape drives must be streamed. That is, they generally need to be sent a stream of data at close to their maximum throughput rate in order to consistently operate well. If they receive a stream of data slower than that, they actually can write slower than the incoming data stream and can also fail more often. See the "Tape Drives" section later in this chapter for more details on this concept.

It may seem silly to say it, but tape drives can really only do two things. They can write data sequentially and read data sequentially. This means that they can be used only with applications that understand how to read and write to tape drivesin other words, backup and archive applications.

In addition, if you've got five backup jobs, you need five tape drives. Yes, many commercial backup software products can interleave onto a single drive jobs that are sent to a single backup server. However, they cannot interleave jobs when they're sent to multiple backup servers. Multiple backup servers often can "time-share" drives, as long as they make sure that no two servers write to the same drive at the same time. However, they cannot actually read and write to the same tape drive simultaneously.

9.1.4.2. Optical drives: A little more flexible

Optical drives are more flexible than tape drives, but aren't quite as flexible as disk drives. They can easily handle different incoming data rates, but they still tend to have the single-minded issue that tape drives have.

Optical drives are random-access devices, which means that they don't have the issues that tape drives have with streaming. Generally, they can handle any data rate up to their native throughput rate. As of this writing, no optical drives use hardware compression, so native throughput rate is the same as effective throughput rate for optical drives.

Since they behave more like disk drives than tape drives, optical drives can also usually be mounted as a filesystem, which means that they also don't have the single-minded issue that tape drives have. Multiple applications can read and write simultaneously to the same optical drive, just like they can to any mounted filesystem. Global filesystem software could also be used with an optical drive to allow multiple servers to simultaneously read and write to it as well.

Optical drives could be made to be as flexible as disk. For example, someone could write drivers that would make an optical drive look like a tape drive. As of this writing, no one is doing that.

9.1.4.3. Disk drives: Very flexible

Disk drives really win in the flexibility category. They can handle any data rate you send them, even going beyond their native throughput, through the use of RAID. They can emulate just about any kind of backup device you want, and they can emulate as many of those devices as you would like.

In addition to being able to handle slower data rates, disk drives can be striped together using RAID to handle very large incoming data rates. A single RAID array can handle hundreds of megabytes per second, or it can handle 1 Kbpswhatever you need.

Thanks to virtualization drivers, disks can also emulate other things. RAID is really several disks emulating one big disk. Virtual tape software allows disks to emulate tape. With virtual tape software, you see a disk array, but your backup server sees a tape drive or a tape library. This is referred to as a virtual tape library, or VTL.

Not only can disk drives emulate tape drives, they can emulate a lot of tape drives. Virtual tape software allows a single disk array to emulate as many tape drives or tape libraries as you like, allowing several servers to easily share the same resource without using anything like global filesystem software. Unlike real tape drives, you can make more virtual tape drives with the push of a button.

There's another interesting way that disk drives are flexible. If your backup software supports multiplexing, and you choose to use that feature when backing up to a virtual tape drive, they can help mitigate the side effects of multiplexing by going significantly faster than a normal tape drive.

Normally, you would not multiplex to a virtual tape drive. The reason you multiplexed to tape was to stream the tape; this is not necessary with a virtual tape drive. However, there are some licensing issues with some VTLs and some backup software packages that may give you reason to multiplex to a VTL. The point of this section is to illustrate that if you chose to multiplex to a virtual tape, it might not impact your restore performance as much as multiplexing to physical tape.

Here's how it works. Suppose your backup software multiplexed several backups to a single tape, and you ask it to restore just one of those backups. What it actually does is read all the multiplexed backups and throw into the bit bucket all the bits it doesn't need. If your tape drive can go only 50 MBps, it means that a significant portion (75 percent with a multiplexing setting of four) of that 50 MBps is being spent reading data that's being thrown away. However, what if you had a "tape drive" that could read data at 150 MBps? If you threw away 75 percent of that, you'd still have 37.5 MBps, which is a respectable transfer rate.

Finally, the truth is that the virtual tape interface is just thatan interface to the disk drive. In the end, the data is still lying on disk, which leaves room for a number of possibilities that are covered in more detail later in the chapter. They include de-duplication, replication, content-awareness, and re-presentation.

9.1.5. Time-to-Data

Transfer speed is not always the most important deciding factor. (In fact, it's starting to be one of the last things I personally look at.) Obviously, having a faster drive makes backing up and restoring large amounts of data easier. However, many restore requests are for a single file or a small group of files. Depending on your environment, this may account for 99 percent of your restores. In addition, many large restores require anything from several tape mounts to thousands of them. (I know of one restore that required mounting 20,000 tapes.) What counts most in these restores is time-to-data: how long does it take to load a volume, seek to the appropriate place on the volume, and start reading data? Streaming tape drives usually are "slower on the draw" and therefore have a much longer time-to-data. The winners in the time-to-data category used to be optical drives. The worst time-to-data value for an optical drive is typically around 12 seconds, allowing 5 seconds for a robotic exchange. If the file being restored is on a platter that is already loaded, time-to-data is less than a second. However, the new winner in this category is the disk drive. With access times in the nanoseconds, it's hard to beat a disk driveeven if the disk drive is emulating a tape drive. How long does it take to rewind, eject, or load a virtual tape?

HSM applications place the highest importance on the time-to-data value because of how HSM works. An unused file is automatically migrated from disk to a less expensive storage medium. The user does not realize this and may eventually attempt to access the file. The HSM system detects that the file is needed and automatically retrieves it from the backup medium. All the user sees is a delay in accessing the file. If the delay is 12 seconds or less, the user may not even notice. However, suppose that the file is placed on a tape drive whose time-to-data is two minutes. A typical user is either going to call the help desk or reboot by that time. That is why HSM environments should use either optical media or one of the newer tape drives that have been designed with very low time-to-data values.

In extremely high-volume HSM environments, it also matters how long it takes to rewind and eject the volume. The time-to-data is added to the data-transfer time and rewind-and-eject time to create something called cycle time. If the cycle time of a particular tape drive is 2 minutes, the HSM system can service 30 file migration requests per hour per drive. An 8-drive autoloader would increase that to 240 files per hour. In comparison, an average optical platter has a cycle time of less than 30 seconds. This means an 8-drive autoloader using optical platters can service up to 1,000 migration requests per hour. Consider carefully how important the time-to-data value is in your environment.

9.1.6. Capacity

Capacity used to be the most important factor when considering a backup drive. It still is in many nonautomated environments. If a full backup of your entire system can fit on one volume, you do not have to worry about swapping volumes in the middle of the night. However, using an autoloader significantly reduces the importance of volume capacity. In fact, having your data on multiple smaller volumes may actually make a restore go faster with today's backup software.

Having a drive of insufficient capacity affects your overall transfer rate, though. Suppose that you are backing up to a 100 GB volume with a transfer rate of 100 MBps and a cycle time of 4 minutes. At 100 MBps, you can fill the entire volume in a little over 16 minutes. You would need to swap volumes, which would take 4 minutes. That means that it actually took 20 minutes to back up the 100 GB, reducing your overall effective throughput to 80 MBpsnot 100. Suppose that there was a bank of 8 of these drives in an autoloader; the overall throughput for an 8-hour period would be reduced from 29 TB to 23 TBa 6 TB difference. If your volume had a capacity of 500 GB, though, the amount of transferred data lost due to volume swaps is reduced to around 1 TB.

Also remember that larger capacity usually means a longer time-to-dataespecially in tape drives. (Obviously, it takes longer to get halfway through a 2,000-foot tape than it does to get halfway through a 1,000-foot tape.) Some tape drives mitigate this by mounting midpoint, but it's still a long way to the end of the tapeeven if you start in the middle.

Another capacity consideration is the intended use of the drive. Are you planning to use backup software that writes continuously to each volume until it is full or will you make lots of small backups to a single volume? For example, suppose that you regularly make database exports to tape and send them to a partner who imports that data into his database. Do you really need a 400 GB tape to export a 500 MB database? Every export will require an expensive 400 GB tape. It would be more cost effective to use a cheaper, smaller tape or optical drive for such a purpose.

9.1.7. Removability

With the advent of virtual tape, we need to consider this aspect of backup drives. Obviously, tape drives and optical drives are removable, and it is considered one of their great features. As of this writing, removable virtual tape cartridges are also beginning to be available. However, most virtual tape and disk units do not have removable media available.

Many customers are beginning to ask whether or not removability is still a requirement for them. The reason that most people need removability is that they want to be able to send their tapes to an off-site vaulting vendor. What if you could do that without moving tape around? Would you still need the media to be removable? Ask yourself this question when examining the replication features of disk targets, covered later in this chapter.

9.1.8. Cost

Drives that are more reliable and flexible, store more data at a faster rate, and have a smaller time-to-data value usually cost more. The only time when you may get a price break on a reliable, quick drive is if a manufacturer is trying to break into a market that is dominated by another manufacturer. Of course, then you bear the risk of the drive being taken off the market. (There have been a number of dead soldiers over the years.)

Another cost factor to consider is reusability of media. There are several write-once-read-many (WORM) technologies that do not allow you to reuse media. These drives have a definite purpose, such as making bootable CDs or unchangeable archives, but you definitely should consider whether you are allowed to reuse the media when calculating total cost of ownership.

Another thing you should consider is the cost involved in managing the intelligence behind the backup system. If you use tape drives, you need to do a lot more design and management of the system to make sure that the system is writing data to tape in a way that utilizes tape drives well. If you use disk, you hardly have to do that, thus significantly reducing your total cost of ownership.

Make sure you consider all cost factors when comparing the cost of the different media types. This is especially true when considering the cost of your on-site storage system because it probably won't be a standalone tape drive. It most likely will be either a tape library, an optical library, or some kind of disk system. Do not make the mistake of comparing only the price of the media: tape will always win. Remember that when it comes to your on-site storage system, that piece of media is worthless without the tape or optical library it goes in. Start with determining how much storage you need based on how many full and incremental backups you're going to store on-site. Suppose, for example, you have determined that you need 200 TB of on-site storage. Compare, then, the cost of a fully populated 200 TB tape library, a fully populated 200 TB optical library, and a 200 TB disk system. A fully populated tape or optical library has all the tape/optical drives you need, all the slots you need, and enough tapes/platters to fill those slots. A fully populated disk system has all the disks you need, along with any RAID or virtualization software or hardware needed to make it behave the way you want it to behave. (Of course, you have to take into consideration the differences in power requirements between a disk system and a tape system.) When you compare things like this, you might be surprised at the numbers, especially when you factor in the data de-duplication features of disk systems.

9.1.9. Summary

While several drives fit into the middle-of-the-road category, most drives excel in one or more of these eight critical factors. Tape drives have historically excelled in the cost of acquisition area; however, the de-duplication features of disk units are changing that. Different backup drives are better than others in the areas of time-to-data, transfer rate, and flexibility. Optical media obviously provides a quick time-to-data, but they definitely have smaller capacity yet cost more than their tape counterparts. Disk drives can be more expensive than either tape or optical media, but not necessarily. They excel in every other decision factor. You must decide which factors are most important in your decision.