Disk-Based Mirroring Is A Lot More Than Just Safe.

Disk-based mirroring is often the first line of defense for speeding the return to operational normalcy following data center outages. Yet today, this powerful, flexible technology can also be used proactively, to perform tasks that can dramatically increase the business value of the enterprise. Before examining just a few of the many innovative applications to which users are leveraging their disk-based mirroring investments in new and creative areas, let's take a look at the technology.

True disk-based mirroring uses the intelligence of the disk subsystem to make mirror images of the data residing on its surfaces. Modern disk subsystems have built-in microprocessors, cache, sophisticated microcode (also known as firmware), and software that can be employed to complete, indeed enhance, the task of writing and reading data to and from the disk or disk array.

Today, more and more functions are being moved from the domain of the host server or mainframe to the intelligent storage subsystem. Sophisticated storage-oriented software handles mapping functions and catalog issues so that the host need not encounter duplicate addresses, to take one example. Users can creatively employ the technology because the second copy of data can be made uniquely addressable to the host. Mirrors can be local or remotely located, that is, inside or outside the same disk subsystems. When outside the same disk subsystem, storage-oriented software is used to move the data from the source to the target location.

Disasfer Recovery

Disaster Recovery (DR) operations can be facilitated with disk-based mirrors. In a traditional DR scenario, backup tapes are shipped offsite via couriers to stand-by locations that hopefully are far enough away not to be affected by the original disaster. Unfortunately, these tapes are already out of date by hours or days because they were run previously as full backups or as incremental backups some time ago. Tapes are subject to human errors such as running them in the wrong sequence or even losing them during transit. Problems might also arise if tape degradation prevents a restore operation after a period of time. In contrast, disk-based mirrors can be established remotely and kept current with up-to-the-last-transaction data. In the event of a disaster, the remote mirror assumes the primary disk role since its data is current. Fast post-disaster restart is the result.

Data Vaulting 

A data vault is an offsite repository of information located across campus or across hundreds or thousands of miles. Its purpose is to store and protect data for possible future use. Innovative companies are vaulting their information on RAID protected disks rather than old fashioned tape that is not protected and worse, is susceptible to considerable amounts of human and media error (Fig 1). Depending on the application, the mirror can be kept as current as the last transaction.

Data vaulting differs from Disaster Recovery in that a DR site usually has a standby computer ready to assume a processing role, usually during the restore activity. A data vault may not have such a host and may even be run in an entirely unattended highly automated site.

Point In Time Backups

By far, the most common use of disk-based mirroring is online simultaneous point in time backups (Fig 2). Users make mirrors of their production data at a particular point in time. The process of making the mirror is very fast. Once the mirror is established, the backup routine is run against the mirror image rather than the production image. Better, this backup process occurs in parallel with production processing.

Consider that backup routines occur in virtually every computing operation every day and that most of them require that the production system be taken offline until completion of the backup. By any measure, putting systems back into production immediately is a productivity enhancement of huge significance, not just for the firm, but for the Gross Domestic Product (GDP) of the nation and, all the while, organizational information is being protected by a strictly followed backup discipline.

Data Warehouse Loading

Data warehouses often contain information about customers: their buying patterns, credit histories, demographics, product preferences, and so on. In the new electronic world of commerce, this information is vital to companies seeking to attain competitive advantage by knowing their customers better. The fresher this information is the more reliable it is and, therefore, decisions based on that data are likely to be more accurate.

The process of updating a data warehouse is called "loading" or "refreshing." Unfortunately, all too often, data warehouse loading is a batch activity that is enormously time consuming. The process is conducted offline--taking down the production system until the batch update is completed--because it takes so long.

It is not surprising that the load is often deferred until a more convenient processing time. On the other hand, if users took advantage of quickly established disk-based mirrors, those mirrors could be used as the source from which to load the data warehouse. Data warehouses would be loaded from the mirror rather than the production volume. Simultaneously, production operations could continue operating in parallel. Updates could be conducted nightly instead of every six weeks, resulting in fresher data and better decisions.

Testing

There are several different types of testing that can benefit from running off a mirror of the data rather than its original. These include performance testing, feature/function testing, software maintenance, new software releases/versions updating, and, of course, Year 2000 testing. All have one thing in common. Since testing is both iterative and destructive, its primary purpose is to discover errors and problems, duplicate them, and then fix them. It would clearly be unwise to test in the production environment. Downtime would be unacceptable. However, since disk-based mirrors can have unique addresses, testing and production can be co-located. If testing with a full image of the production environment, issues of scalability disappear.

Really intelligent disk-based mirroring schemes use techniques to trace changes that have occurred causing the original and the mirror to differ. They, then, synchronize the two by conveying the changed data only. In a testing environment, this technique is extremely important because it can mean that the time to refresh the test environment back to its original pristine condition is short. Such resynchronizing takes minutes instead of days, as would occur if using a traditional tape-based scheme with test data spread throughout the production system.

Database Integrity Checks

Most database management systems include tools to insure the integrity of the data they control. Unfortunately, all too often, users do not run these database integrity routines because they are either resource intensive or they take too long. Both excuses put the organization at risk. Omitting database integrity checks can lead to corrupted data residing inside the database. Depending on how long it takes before the error is discovered (minutes, hours, and even days), the organization may suffer increasing financial repercussions.

Alternatively, a uniquely addressable mirror of the database could be split off. Using a different server or LPAR (logical partition), the validity of the database could be insured as a result of a background processing task against the mirror without interrupting the production usage of the source data. Once verification is completed, the whole process could begin anew in a never-ending proof cycle. If corruption ever does occur, the system can return to the last set of data proven correct.

Data Replication And Data Distribution

Electronic commerce is leading the way to using disk-based mirrors for data replication and data distribution. If it is important to a new e-firm to have the same data synchronized at two or more different locations, then disk-based mirroring can be employed. For example, maybe an Internet Service Provider (ISP) makes its profits from banner advertising on its Web pages. If that firm wants to have service points on the East Coast, in the Midwest, in the mountains region, and on the West Coast to lower the transmission times for its subscribers, it may be important that all four locations have the same advertising data. The ISP would use disk-based mirrors to achieve the data replication automatically.

Alternatively, perhaps each office of a major company needs to have nightly updates of its "pricing" files from corporate headquarters. In this case, the master price file could have a mirror established. That mirror would, then, be the source document for distributions to all branch locations. While transmissions were taking place between the host and the branches using storage based mirroring software, the mainframe could be put back online to handle incoming market orders from time zones around the electronic world.

While disk mirroring is often considered a "safety" process, to be used only to protect data, this view looks to the past, not toward the future. Mirroring has finally reached a level of sophistication that allows the technology to be used in numerous ways, to increase productivity, and make information available to anyone at any time.

James W. Baker is a senior product manager at EMC Corporation (Hopkinton, MA).