The first article in a three-part series examines the pros and cons of the three approaches to replication.
By Karen Dutch
It should come as no surprise that replication is a hot topic today. Because of the huge increase in the amount of mission-critical data—as well as growing concerns and new regulations regarding data availability—disaster recovery and business continuance are leading IT managers to revisit their current data-protection and recovery plans. As a result, they are looking at adopting data replication as an essential component of disaster protection to enhance current strategies.
Replication is the process of making an exact copy of an organization's current data, without disrupting or disabling active applications. The copy may be kept on an adjacent system, ready to use in an instant should the primary system fail, or it may reside at a distant location, quickly available should a disaster hit the primary data center, knocking it out of service. In addition, the copies may be used for application testing or development, as well as for offloading resource-intensive reporting.
In the past, replication was relatively expensive and was only used for the most-mission-critical data. Replication required a duplicate set of high-end, expensive storage systems and sufficient network bandwidth. Only the largest companies could afford this kind of duplication.
Today, however, due to less-expensive storage (such as ATA and Serial ATA disk arrays), servers, and bandwidth, as well as new ways to copy data continuously over long distances, companies are looking at implementing more cost-effective replication solutions. Additionally, there is more data that is now considered mission-critical. Replication is becoming more necessary as an element of disaster recovery and is also now practical even for small and mid-sized companies.
But the immediate question is, "Which type of replication should you use?" Replication is available in a number of types, each offering strengths and limitations. Before organizations make a replication purchase decision, they need to understand the pros and cons of each option.
Briefly, replication can be either mirrors (synchronous or asynchronous) or point-in-time copies. Synchronous and asynchronous replication maintains data consistency both within and across volumes and is used to mitigate against data loss due to physical disruptions (e.g., disk, network, system, or site failures).
With synchronous replication, both the source and target devices acknowledge a write before the next write can occur. Synchronous replication is primarily used in high-availability situations where in the event of a failure of the primary system a reserve system needs to take over immediately, picking up at the exact point of failure. This type of replication is traditionally used over short distances so that application performance is not impacted.
Asynchronous replication allows writes to continue to the source volumes while the target volumes are updated in the background. It is used primarily in disaster-recovery scenarios where the recovery site is located far away and the application would experience severe performance degradation with synchronous replication.
Point-in-time copy, sometimes referred to as a snapshot, is a copy of data taken at a specific point in time. Point-in-time copies can also span one or more volumes, and most replication tools can create multiple copies for source storage.
Point-in-time copies enable organizations to recover from logical disruptions such as data corruption, user error, or virus attacks. A point-in-time copy restores data to a known stable state prior to the event that caused the disruption. Point-in-time copies can also be used to complement traditional backup processes. Companies should consider combinations of both mirroring and point-in-time replication as part of an overall data availability and protection plan.
Three types of replication
Today, IT managers can choose among three types of replication: host-based, network-based, and array-based. Which type of replication the organization chooses should depend on its specific replication objectives, budget, and storage environment.
Host-based replication runs on both the source and target and is in the data path. The data is replicated at a logical level rather than at the physical level. Because the data is replicated at the logical level, the primary and secondary storage devices do not need to have the same physical characteristics or come from the same vendor.
Network-based replication uses a switch or appliance that sits on the network in the data path to handle replication. Network-based replication devices have proprietary replication technology, but are vendor-agnostic across host servers and storage hardware.
Array-based replication was the first method available and is provided by the major hardware vendors. It was initially available only for high-end storage devices, but is now becoming available on mid-tier arrays. However, this method only replicates data between storage controllers from the same vendor. While array-based replication requires proprietary hardware, it is not operating-system or server-dependent.
Pros and cons
Each replication approach has strengths and limitations. For example, host-based replication is flexible in that it can leverage existing IP networks. It supports data replication between any two storage arrays and does not require any additional hardware to replicate data. However, host-based replication can be more difficult to manage with large groups of servers if the tool does not provide a centralized management console. In addition, it consumes host resources during replication and has operating system dependencies.
Network-based replication removes the overhead from the application server and is particularly effective in large, heterogeneous storage and server environments. However, network-based replication requires additional proprietary hardware, which may represent a higher initial cost, and may require implementation of a storage area networks (SAN). In contrast, host-based replication works in a SAN or direct-attached storage (DAS) environment.
Array-based replication offers high performance and optimizes host server resources by offloading replication to array controllers. However, array-based replication requires homogeneous storage and dedicated network pipes between the arrays, making it a high-cost solution. In addition, setup and management of array-based replication can be difficult and error-prone.
Each type of replication also has cost implications. Host-based replication, for example, drains host resources, which can impact performance and require host upgrades or the purchase of a larger host. Network-based replication requires the acquisition of another device or appliance, which entails acquisition costs and ongoing operational and management costs. Array-based replication may require the acquisition of external equipment to convert disk channel protocols for use with WANs, and vendor lock-in hinders attempts to lower storage acquisition costs in the future.
There is no single correct replication choice. IT managers must carefully assess their storage and server environments in light of their replication needs, strategy, and budget. With three good choices, however, there is no longer any reason for an IT organization to do without replication, which is the only way to truly protect the organization's critical data and ensure its availability throughout the enterprise.
In the next article, we will look at how to use replication to improve a disaster-recovery plan. A subsequent article will examine a range of other uses for replicated data that leverage the value of the organization's investment in data replication.
Karen Dutch is vice president of product management at Fujitsu Softek (www.softek.fujitsu.com) in Sunnyvale, CA.