Assessing the foundation of long distance disaster recovery.

In theory, enterprise data protection strategies should be geared towards enabling the rapid recovery of business operations in the event of a wide range of failures. The continuum of failure types spans the range from a simple hardware component failure to a widespread disaster that may take out one or more data centers. Current recovery strategies such as RAID or other redundant components, I/O multi-pathing and local clustering for high availability purposes do a fairly good job of supporting rapid and, in some cases, even transparent recovery from simple local failures. But more widespread types of failures (due to acts of terrorism, forces of nature such as tornadoes or earthquakes, or events such as multi-state power outages) raise the bar significantly to operational recovery.

All recovery options must start with the data. The obvious solution to the problem of widespread disasters has been to locate one or more copies of the operational data at one or more sites located far enough apart so as not to be affected by any single disaster. Although there are a variety of other issues that must be addressed to ensure rapid and reliable recover--such as human resources, proven processes, communications and access to appropriate equipment--the foundation for viable recovery start with the data. Following is what is required to lay the foundation for enterprise data recoverability.

Current Long Distance Disaster Recovery Solutions

Traditionally, there have been two approaches to this problem. Tape-based backup solutions store a point-in-time copy of operational data on inexpensive media that can be moved to a remote location and safely stored. If recovery is required, data can be restored from tape to disk at a recovery site and then used to restart critical applications. The other approach has been to use asynchronous replication to continuously maintain a relatively up-to-date copy of operational data on disk at a remote site. Any changes to operational data at the local site are sent across a network of some type to also be applied at the remote site. If recovery is required, servers and applications at the remote site can be brought up using this copy.

Each of these solutions introduces major operational trade offs, however. Tape-based backup solutions suffer from three basic problems: backup windows, data loss and data integrity/recoverability.

[FIGURE 1 OMITTED]

The backup window is the amount of time an application must be off line for a backup to be performed. In today's globalized environment, 7X24 availability is a strong requirement, leaving literally no time when critical applications can be down--even for just a minute or two. Even the use of snapshot technology still requires the application to be brought down, causing operational impact.

Tape-based backups are also point-in-time copies, current to the point when the backup was taken. Given the backup window problem, backups are not taken very frequently. Most environments are backed up only once a day at the most, and many are only backed up every several days or on a weekly basis. If recovery is required, any changes to the operational data made since the last backup are lost.

Finally, tape media integrity can introduce non-deterministic recovery problems. There is some risk in the data conversion that takes place when data is backed up from disk to tape, and then restored from tape back to disk. Tapes must be managed with some manual labor at some point throughout the off-site storage process, and this introduces the possibility of human error (tapes could be misplaced or lost). Tapes also wear out more quickly than disks, particularly if they are being used over and over for backup purposes. Research done by Gartner Group in 2H03 indicates that one in four tapes have unrecoverable files. The unfortunate aspect of this is that it is not clear if a file is unrecoverable until recovery is attempted, and at that point it's too late to do anything about it if it can't be recovered.

Replication did a good job of addressing these issues. Once installed, replication can run continuously in the background, effectively removing the backup window issue. Data loss is minimized relative to tape since writes are continuously being sent across the network to the remote site. In almost all cases, replication will allow recovery from much more current data than tape. And because data at the remote site is stored on disk in native disk format (not converted to a tape format for storage on tape) there is a much higher chance of data being recoverable when it is required.

Historically, replication has been significantly more expensive than tape due to the price differential between disk and tape hardware. But newer disk technologies such as ATA are closing the cost gap while at the same time providing more reliable solutions with faster recovery. The disk-to-disk data protection trend is definitely taking hold. In December 2003, an Enterprise Storage Group survey showed that 83% of enterprise users and 59% of mid-tier users have either already deployed or state that they will purchase some form of disk-based data protection technology within the next 24 months. Replication is one form of disk-to-disk data protection that will benefit from this trend.

The first form of replication available was synchronous replication. In synchronous replication, widely deployed for sites within 30-50 miles of each other, writes at both sites must complete before a write acknowledgement is sent back to the critical application(s). The write latency between sites then becomes a constraint limiting the viable distance of synchronous replication configurations. To address the concerns over widespread disasters in this post-9/11 era, synchronous replication is not sufficient, in most cases, because of distance limitations.

By its very nature, asynchronous replication is designed for long-distance configurations. In asynchronous replication, the write to the remote site is decoupled from the write to the local site using a local queue of some sort. This means that critical application performance is unfettered by the distance between sites, opening up the ability to support configurations spanning literally thousands of miles. One key point to understand about asynchronous replication is that the asynchronous nature of the write does mean that there can be some lag time between writes being applied at the local and remote sites, depending on network bandwidth and distance. This lag time is generally measured on the order of seconds or minutes, however, as opposed to tape where the "lag" time (the last backup point) is often measured in days.

Available Implementations

Asynchronous replication is available in a variety of forms. Storage-based solutions enable asynchronous replication between enterprise arrays from the same vendor and can support heterogeneous servers, but generally force vendor lock-in at the array level, driving high cost. Server-based solutions support mirroring between servers running the same operating system and can support heterogeneous arrays, but can also impose vendor lock-in at the volume manager or file system level. Depending on the vendor, architectural limitations may also impose performance and scalability issues. Appliance-based solutions support both heterogeneous servers and storage, but force new hardware purchases (appliances) and can introduce performance bottlenecks that limit scalability at the appliance level. Generally, all of these types of solutions support the ability to replicate data over IP-based networks. If long distance replication is required, it is important to understand the features and limitations of each approach to determine if it is appropriate for your environment.

There are vendors currently developing network device-based solutions that are slated for availability in 2005. These solutions are expected to offer an operating system-agnostic solution with no requirement for any host-based components. The availability of switch-based options will offer users increased choice in how they deploy replication.

Key considerations when evaluating an asynchronous replication solution include:

* Does it support heterogeneous solutions that meet your definition of "heterogeneous"? Is ATA and other cost-effective disk supported?

* Does it have the flexibility to support multiple storage architectures (DAS, SAN and NAS), preserve existing investments and leave future investment options open?

* Does it impose any form of vendor lock-in at the operating system, volume manager, file system, application, server or storage level?

* What are the data integrity and scalability limitations (if any) of the architectural implementation?

* Does the vendor offer some way to evaluate the performance of their solution in your real environment prior to purchase?

* What are the operational implications, including cost, of installing the solution in your existing environment?

Asynchronous replication can be an integral part of an overall disaster recovery solution that needs to solve the "widespread disaster" problem.

Understanding the capabilities of asynchronous replication will ensure it can be most effectively applied in a given environment, providing the best solution for the money.

Eric Burgener is vice president of marketing at Topio (Santa Clara, CA)

www.topio.com