Backups

tar

The tar utility is well known among UNIX system administrators. It is the archiving method of choice for sharing ad-hoc bits of source code and files between systems. The tar implementation included with Red Hat Linux is GNU tar, one of the more feature-rich tar implementations.

Backing up the contents of a directory can be as simple as issuing a command similar to the following:

tar cf /mnt/backup/home-backup.tar /home/

This command will create an archive called home-backup.tar in /mnt/backup/. The archive will contain the contents of the /home/ directory. The archive file can be compressed by adding a single option:

tar czf /mnt/backup/home-backup.tar.gz /home/

The home-backup.tar.gz file is now gzip compressed.

There are many other options to tar; to learn more about them, read the tar man page.

cpio

The cpio utility is another traditional UNIX program. It is an excellent general-purpose program for moving data from one place to another and, as such, can serve well as a backup program.

The behavior of cpio is a bit different from tar. Unlike tar, cpio reads the files it is to process via standard input. A common method of generating a list of files for cpio is to use programs such as find whose output is then piped to cpio:

find /home | cpio -o > /mnt/backup/home-backup.cpio

This command creates a cpio archive called home-backup.cpio in the /mnt/backup directory.

There are many other options to cpio; to learn more about them see the cpio man page.

dump/restore: Not Recommended!

The dump and restore programs are Linux equivalents to the UNIX programs of the same name. As such, many system administrators with UNIX experience may feel that dump and restore are viable candidates for a good backup program under Red Hat Linux. Unfortunately, the design of the Linux kernel has moved ahead of dump's design. Here is Linus Torvald's comment on the subject:

From:	 Linus Torvalds
To:	 Neil Conway
Subject: Re: [PATCH] SMP race in ext2 - metadata corruption.
Date:	 Fri, 27 Apr 2001 09:59:46 -0700 (PDT)
Cc:	 Kernel Mailing List <linux-kernel At vger Dot kernel Dot org>

[ linux-kernel added back as a cc ]

On Fri, 27 Apr 2001, Neil Conway wrote:
> > I'm surprised that dump is deprecated (by you at least ;-)).  What to
> use instead for backups on machines that can't umount disks regularly? 


Note that dump simply won't work reliably at all even in 2.4.x: the buffer
cache and the page cache (where all the actual data is) are not
coherent. This is only going to get even worse in 2.5.x, when the
directories are moved into the page cache as well.

So anybody who depends on "dump" getting backups right is already playing
Russian roulette with their backups.  It's not at all guaranteed to get the
right results - you may end up having stale data in the buffer cache that
ends up being "backed up".

Dump was a stupid program in the first place. Leave it behind.

> I've always thought "tar" was a bit undesirable (updates atimes or
> ctimes for example).

Right now, the cpio/tar/xxx solutions are definitely the best ones, and
will work on multiple filesystems (another limitation of "dump"). Whatever
problems they have, they are still better than the _guaranteed_(*)  data
corruptions of "dump".

However, it may be that in the long run it would be advantageous to have a
"filesystem maintenance interface" for doing things like backups and
defragmentation..

		Linus

(*) Dump may work fine for you a thousand times. But it _will_ fail under
the right circumstances. And there is nothing you can do about it.

Given this problem, the use of dump/restore is strongly discouraged.

The Advanced Maryland Automatic Network Disk Archiver (AMANDA)

AMANDA is a client/server based backup application produced by the University of Maryland. By having a client/server architecture, a single backup server (normally a fairly powerful system with a great deal of free space on fast disks, and configured with the desired backup device) can back up many client systems, which need nothing more than the AMANDA client software.

This approach to backups makes a great deal of sense, as it concentrates those resources needed for backups in one system, instead of requiring additional hardware for every system requiring backup services. AMANDA's design also serves to centralize the administration of backups, making the system administrator's life that much easier.

The AMANDA server manages a pool of backup media, and rotates usage through the pool in order to ensure that all backups are retained for the administrator-dictated timeframe. All media is pre-formatted with data that allows AMANDA to detect whether the proper media is available or not. In addition, AMANDA can be interfaced with robotic media changing units, making it possible to completely automate backups.

AMANDA can use either tar or dump to do the actual backups (although under Red Hat Linux using tar is preferable, due to the issues with dump raised in the Section called dump/restore: Not Recommended!). As such, AMANDA backups do not require AMANDA in order to restore files — a decided plus.

In operation, AMANDA is normally scheduled to run once a day during the data center's backup window. The AMANDA server connects to the client systems, and directs the clients to produce estimated sizes of the backups to be done. Once all the estimates are available, the server constructs a schedule, automatically determining the order in which systems will be backed up.

Once the backups actually start, the data is sent over the network from the client to the server, where it is stored on a holding disk. Once a backup is complete, the server starts writing it out from the holding disk to the backup media. At the same time, other clients are sending their backups to the server for storage on the holding disk. This results in a continuous stream of data available for writing to the backup media. As backups are written to the backup media, they are deleted from the server's holding disk.

Once all backups have been completed, the system administrator is emailed a report outlining the status of the backups, making review easy and fast.

Should it be necessary to restore data, AMANDA contains a utility program that allows the operator to identify the file system, date, and file name(s). Once this is done, AMANDA identifies the correct backup media, accesses, and restores the desired data. As stated earlier, AMANDA's design also makes it possible to restore data even without AMANDA's assistance, although identification of the correct media would be a slower, manual process.

This section has only touched upon the most basic AMANDA concepts. If you would like to do more research on AMANDA, your Red Hat Linux system has additional information. To learn more, type the following command for a list of documentation files available for AMANDA:

rpm -qd amanda-server

(Note that this command will only work if you have installed the amanda RPMs on your Red Hat Linux system.)

You can also learn more about AMANDA from the AMANDA website at http://www.amanda.org/.

Full Backups

The type of backup that was discussed at the beginning of this section is known as a full backup. A full backup is simply a backup where every single file is written to the backup media. As noted above, if the data being backed up never changes, every full backup being created will be the same.

That similarity is due to the fact that a full backup does not check to see if a file has changed since the last backup; it blindly writes it to the backup media whether it has been modified or not.

This is the reason why full backups are not done all the time — every file is written to the backup media. This means that a great deal of backup media is used even if nothing has changed. Backing up 100 gigabytes of data each night when maybe 10 megabytes worth of data has changed is not a sound approach; that is why incremental backups were created.

Incremental Backups

Unlike full backups, incremental backups first look to see whether a file's modification time is more recent than its last backup time. If it is not, that file has not been modified since the last backup and can be skipped this time. On the other hand, if the modification date is more recent than the last backup date, the file has been modified and should be backed up.

Incremental backups are used in conjunction with an occasional full backup (for example, a weekly full backup, with daily incrementals).

The primary advantage gained by using incremental backups is that the incremental backups run more quickly than full backups. The primary disadvantage to incremental backups is that restoring any given file may mean going through one or more incremental backups until the file is found. When restoring a complete file system, it is necessary to restore the last full backup and every subsequent incremental backup.

In an attempt to alleviate the need to go through every incremental backup, a slightly different approach was implemented. This is known as the differential backup.

Differential Backups

Differential backups are similar to incremental backups in that both backup only modified files. However, differential backups are cumulative — in other words, with a differential backup, if a file is modified and backed up on Tuesday night, it will also be backed up on Wednesday night (even if it has not been modified since).

Of course, all newly-modified files will be backed up as well.

Like the backup strategy used with incremental backups, differential backups normally follow the same approach: a single periodic full backup followed by more frequent differential backups.

The affect of using differential backups in this way is that the differential backups tend to grow a bit over time (assuming different files are modified over the time between full backups). However, the benefit to differential backups comes at restoration time — at most, the latest full backup and the latest differential backup will need to be restored.

Tape

Tape was the first widely-used removable data storage medium. It has the benefits of low media cost, and reasonably-good storage capacity. However, tape has some disadvantages — it is subject to wear, and data access on tape is sequential in nature.

These factors mean that it is necessary to keep track of tape usage (retiring tapes once they have reached the end of their useful life), and that searching for a specific file on tape can be a lengthy proposition.

On the other hand, tape is one of the most inexpensive mass storage media available, and it has a long history of reliability. This means that building a good-sized tape library need not consume a large part of your budget, and you can count on it being usable now and in the future.

Disk

In years past, disk drives would never have been used as a backup medium. However, storage prices have dropped to the point where, in some cases, using disk drives for backup storage does make sense.

The primary reason for using disk drives as a backup medium would be speed. There is no faster mass storage medium available. Speed can be a critical factor when your data center's backup window is short, and the amount of data to be backed up is large.

But disk storage is not the ideal backup medium, for a number of reasons:

• Disk drives are not normally removable. One key factor to an effective backup strategy is to get the backups out of your data center and into off-site storage of some sort. A backup of your production database sitting on a disk drive two feet away from the database itself is not a backup; it is a copy. And copies are not very useful should the data center and its contents (including your copies) be damaged or destroyed by some unfortunate set of circumstances.

• Disk drives are expensive (at least compared to other backup media). There may be circumstances where money truly is no object, but in all other circumstances, the expenses associated with using disk drives for backup mean that the number of backup copies will be kept low to keep the overall cost of backups low. Fewer backup copies mean less redundancy should a backup not be readable for some reason.

• Disk drives are fragile. Even if you spend the extra money for removable disk drives, their fragility can be a problem. If you drop a disk drive, you have lost your backup. It is possible to purchase specialized cases that can reduce (but not entirely eliminate) this hazard, but that makes an already-expensive proposition even more so.

• Disk drives are not archival media. Even assuming you are able to overcome all the other problems associated with performing backups onto disk drives, you should consider the following. Most organizations have various legal requirements for keeping records available for certain lengths of time. The chance of getting usable data from a 20-year-old tape is much greater than the chance of getting usable data from a 20-year-old disk drive. For instance, would you still have the hardware necessary to connect it to your system? Another thing to consider is that a disk drive is much more complex than a tape cartridge. When a 20-year-old motor spins a 20-year-old disk platter, causing 20-year-old read/write heads to fly over the platter surface, what are the chances that all these components will work flawlessly after sitting idle for 20 years?

  	Note
  	Some data centers back up to disk drives and then, when the backups have been completed, the backups are written out to tape for archival purposes. In many respects this is similar to how AMANDA handles backups.

All this said, there are still some instances where backing up to disk drives might make sense. In the next section we will see how they can be combined with a network to form a viable backup solution.

Network

By itself, a network cannot act as backup media. But combined with mass storage technologies, it can serve quite well. For instance, by combining a high-speed network link to a remote data center containing large amounts of disk storage, suddenly the disadvantages about backing up to disks mentioned earlier are no longer disadvantages.

By backing up over the network, the disk drives are already off-site, so there is no need for transporting fragile disk drives anywhere. With enough network bandwidth, the speed advantage you can get from disk drives is maintained.

However, this approach still does nothing to address the matter of archival storage (though the same "spin off to tape after the backup" approach mentioned earlier can be used). In addition, the costs of a remote data center with a high-speed link to the main data center make this solution extremely expensive. But for the types of organizations that need the kind of features this solution can provide, it is a cost they will gladly pay.

Restoring From the Bare Metal

The phrase "restoring from the bare metal" is system administrator's way of describing the process of restoring a complete system backup onto a computer with absolutely no data of any kind on it — no operating system, no applications, nothing.

Although some computers have the ability to create bootable backup tapes, and to actually boot from them to start the restoration process, the PC architecture used in most systems running Red Hat Linux do not lend themselves to this approach. However, some alternatives are available:

Red Hat Linux supports both of these approaches. In order to be prepared, you should try a bare metal restore from time to time (and especially whenever there has been any significant change in the system environment).

Testing Backups

Every type of backup should be tested on a periodic basis to make sure that data can be read from it. It is a fact that sometimes backups are performed that are, for one reason or another, unreadable. The unfortunate part in all this is that many times it is not realized until data has been lost and must be restored from backup.

The reasons for this can range from changes in tape drive head alignment, misconfigured backup software, and operator error. No matter what the cause, without periodic testing you cannot be sure that you are actually generating backups from which data can be restored at some later time.