DRJ's Spring 2019

Conference & Exhibit

Attend The #1 BC/DR Event!

Winter Journal

Volume 31, Issue 4

Full Contents Now Available!


This article is about replication of information, not just disaster recovery, and how it can help improve business continuance for users of Information Technology.

While many consider replication primarily a solution for disaster recovery, the reality is that using a robust replication scheme reduces and even eliminates outages in data availability - planned and unplanned. Reducing or eliminating planned outages means mission-critical applications go on generating revenue and improving a company’s bottom line despite interruptions.

By reducing a company’s exposure to unplanned outages, the company may be able to continue generating revenue even when the worst cataclysm hits one of their data centers.

Why replicate?

It seems obvious but bears repeating: Replication is really not an option but a necessity. Some reasons: online backups, data migration projects, application testing, data mining, storage consolidation, storage upgrades, and disaster recovery - all require data replication.

Replicating allows several tasks to be performed on the second image without affecting the original. For example, the second image may be used as a source when performing a backup. Creating a second copy of the data (potentially on a second system), creates an offline backup without impacting the performance of the primary data set.

A large number of IT professionals distrust online backup. No one wants to slow down performance during normal operation of the database just to create a backup.

And, an application’s performance requirements might have outgrown the capability of the storage platform. To get higher performance or to exploit higher capacity storage platforms calls for replication, as well. Replication can ease the migration task of moving the data from low-end islands of storage from several servers to a highly available high performance centralized storage platform, for example.

In a third case, a “live” database often is needed to test a new software version or verify a database change. Again, the case is made for replication.

Finally, many businesses concerned with potential business interruption replicate their data in a remote data center - across town, across state, across the globe. These users require easy transport of data from the primary facility to the other. In the process, they might also replicate data from the remote site to the primary for a cross-mirroring solution.

Back to the basics

To decide which replication scheme should be used, a user needs to decide which method best fits the needs of the business -- server-, controller-, or appliance-based. And, users must decide what to replicate, data or storage.

Data vs. storage replication

Replicating file- or record-oriented information can be considered data replication because that is exactly what is being replicated - the data.

Replicating file- or record-oriented data typically requires a server-based solution, as hardware-based solutions are not able to understand as to the context of the data being written.

Data replication might be a reasonable choice if there are several different areas or types of data or information being used on a disk or Logical Unit Number (LUN), and only a subset of that information must or should be replicated.

The key is deciding precisely what is to be replicated. For example, if replicating a Unix boot disk it is recommended to not replicate the swap area on the disk. If replicating from a NT system, do not replicate the pagefile.sys file.

In both of these cases, these areas or files are used for the virtual memory of the system. They are not useful on any other system and are constantly being accessed and modified.
In another example, users might decide to replicate a database’s index, roll, and possibly the data files but not the temporary work area.

Another option is to replicate the entire disk or LUN, making no distinction of what is actually being replicated. This can be considered storage replication and involves an entire disk or LUN, as there is no distinction of the type or format of the information being replicated.

Storage replication can be performed by either a software- or hardware-based solution. Storage replication is useful when an entire disk or LUN is being used for a single purpose or all of the data on that disk or LUN requires replication.

With the large datasets being used today, it is certainly not unusual for an entire disk or LUN to be used for a single purpose. In fact, there may be many disks or LUNs used for a single purpose or database and typically they would all have to be replicated in order to be useful.

Mirrors and copies

A copy of data is a stand-alone image of data at a single point in time.
Mirrors are copies of data that may be updated as changes are made to the primary instance of the data.

Copies may be used for back-ups or data forms, where mirrors would typically be used for disaster recovery.

Full Mirrors

A full mirror contains a complete copy of the data that uses the same amount of space on the secondary image as the primary.

In other words, a full mirror of a 100GB dataset would take another 100GB - for a total of 200GB.

If RAID is being used for the storage, make the calculations on the available storage and not the raw storage.

Using the same example of a 100GB dataset. If that dataset needs to have 100GB of available storage using a RAID-1 for both the primary and secondary of a mirror pair -400GB of raw storage is actually needed to get the 200GB of usable storage (2 images of 100GB each, consisting of 200GB raw storage each).

If using RAID-5 or RAID-3, naturally the multiplier for usable to raw storage will change. To keep things simple, it is best to estimate the amount of usable storage required and then multiply that number out to for the RAID implementation and then double that number for a single mirror.

Full mirrors will generally have five states: establishing, established, suspended, terminated and re-establishing. When the mirrors are establishing, the original data is being read from the primary image and written to the secondary, or mirror, image.

When the mirrors are established, the images are identical and writes are sent to both images simultaneously (at least logically).

Suspended mirrors allow for the two images to be split for a time, and writes made to one image are tracked for use during the re-establish process. Depending on the mirroring implementation, there may be restrictions as to what can be suspended and what cannot, such as only the secondary or either the primary or the secondary.

Terminated mirrors means just that: the mirror pair is broken and each image can be separately addressed and written to; there is no longer a relationship between the two images.

Re-establishing the mirrors is a state where the two images are resynchronized by sending the outstanding changes to the previously suspended image.

The method of tracking the changes and how they may be written is implementation specific. A full mirror is especially useful in disaster recovery and business continuance models, storage migration projects, and software upgrade testing. This is especially true when it is easy to select which image to fall back to during the resynchronization process.

Flash mirrors

Some mirrors or copies of data are available instantly, in a flash, where others may not be usable until the establish or copy process has been completed. Until the images are synchronized, writes may actually involve two writes and a read: one read to grab the old data, one write to send the old data to the flash mirror, and one write of the new data to the live data image.

In addition, a background task will fill up the blanks on the mirror image. The advantage of this is that work can be done on the flashed image without waiting for the image to be completely copied over.

When images are hundreds of GB in size, there may be a significant time saving.

On the other hand, there may be a significant amount of additional overhead in supporting this model because of the additional read/write traffic.

Flash mirrors are useful when write performance is not as critical as making the mirrors available quickly, such as using a point-in-time copy that is regularly reset to reflect the current state every day or so.

Sparse copies

A sparse copy works on the theory that while it exists; it only needs to keep track of the data that has been over-written.

These are typically point in time images that have a limited life expectancy (such as while the data is being backed up). Instead of using an identically sized image as what was used in the full mirror, a separate area is used that only needs to be roughly the size of the amount of data being written while it is in existence.

For example, a 100GB database may only write 1GB of data during the time a backup takes place. In this example, the sparse mirror needs to be only 1GB in size instead of the 100 GB size a full mirror would require.

A sparse copy is created by assigning a device (or space on a device or file system) to be used as the sparse storage repository. Any writes after this point in time are then written to the primary image after the old data is saved in the repository (similar in some respects to the flash mirror).

The difference here is that data written is not saved on a block-for-block basis as in the case of the full mirror. Instead the data writes are sent to the sparse repository starting at the beginning of the sparse repository, working its way towards the end. Some method is used to track which data blocks have been written to the repository, so that any subsequent read would know to read either from the primary or where to look on the sparse repository (depending on the image being read and whether the data has been modified or not).

In some implementations, there may be multiple images kept in the sparse repository, and it depends on the implementation as to how multiple writes to the same area are handled.

There are two problems with a sparse copy: if the primary image develops a problem, such as a device goes off-line, all of the images become unavailable; and if the amount of data being written exceeds the capacity of the sparse data repository the sparse copy is no longer valid.

On the other hand, if the data is only temporary and has a limited life expectancy, there is a significant amount of storage that can be saved compared to a full mirror.


There are a few very critical decisions to be made at the outset to insure that data replication succeeds and the full benefits are realized.

Each decision grows out of prior decisions. And, like the bricks in a wall, each decision leads to a structure that meets the business needs of the organization in the most time-, cost- and resource-saving way.

Decision then leads to implementation.*

Robert A. Collar is Senior Product Manager for SAN Director Products at LSI Logic Storage Systems, Inc., Milpitas, CA. He has been involved in high availability solutions off and on since 1988, working at Tolerant System, Pyramid, and as an IBM/HP/SGI reseller. He has been involved in Unix-based solutions since 1979. He recently addressed the SNW/Tokyo show in January 2001 on replication and business continuance. He can be reached at This email address is being protected from spambots. You need JavaScript enabled to view it..
*Part 2 of this article will be featured in the next issue of the DRJ.