Wednesday, December 1, 2010

Change-based replication can make data movement between the primary data center and a DR site economically feasible but the process can still involve a significant amount of complexity and infrastructure. The remote replication process itself can be done at a number of points in the environment: at the application level with software running on application servers; with a dedicated appliance sitting in front of the primary storage system; as a software function within that storage array or by using backup appliances. In most cases these options require similar or identical hardware and software at the DR site and they often involve a separate process to maintain and manage. But without real data reduction, (more than simple deduplication) off-site replication can still take too long and/or eat up too much bandwidth, especially during the recovery. To improve performance many customers are forced to purchase WAN accelerating appliances that provide additional compression and deduplication as well as advanced packet shaping. The problem is that these devices also add to the cost of the project and often, to, its complexity.

Next, there are the details of the DR site itself. At the high end is the option of setting up a dedicated data center in a separate geographic location. A more economical alternative may be to use an existing office, but that still requires physical space, power, etc, and some arrangement with onsite personnel to manage the DR equipment. A colocation facility is also an option, but still requires the storage and compute infrastructure.

Cloud storage solutions offer a new option, as they provide a ready-made DR ‘target’ location, an attractive pricing model and near-infinite scalability. But most cloud storage solutions require a separate system on-site to manage the interface and data transfer to the cloud. This represents another ‘box in the data path’, separate from the storage array, performing a dedicated function. Also, restoring from cloud storage has all the same problems that restoring from an off-site DR location does: high bandwidth costs and delays. In the end the effectiveness of the entire process is also related to the amount of data moved. More data means more capacity, more bandwidth, more complexity and more cost. These solutions must do something to reduce the amount of data that’s transferred between the primary and DR storage systems.

There are Hybrid Cloud solutions available now that integrate primary storage with a cloud interface into a single system to make remote replication technology work in a real DR scenario - one that’s cost effective for more than just mission critical applications. The StorSimple appliance provides primary data storage performance with local SSD and SAS drives, optimized with automated tiering, dedupe and application-aware storage volumes. It provides virtually unlimited snapshots and off-site replication, faster snapshot restores and near ‘instant restores’ via remote virtual servers.

Cloud storage can provide an off-site DR target ‘facility’ that’s affordable and scalable; no secondary site, no sharing remote offices, no additional staff allocation, no colocation facilities, etc. Primary storage systems with an integrated cloud interface eliminate the second appliance that cloud storage typically requires, and the management overhead that comes with it. These solutions also eliminate the need for the separate remote replication hardware or software that traditional DR solutions use, simplifying the infrastructure. But they also address the data reduction issue.

Data reduction requires real-time deduplication in the primary storage system, but also some intelligence about the data itself. Hybrid cloud storage systems like StorSimple’s have an application awareness that enables them to differentiate between application-specific data types and data segments (like metadata, BLOBs, transaction logs, etc) and create data volumes specific to each type. This allows for further data reduction as duplicate segments are identified at the application level. Application awareness also enables the identification of a relatively small percentage of the total data, or the most active ‘working set’, which can then be stored on the fastest tiers. On the recovery side, this intelligence enables these most critical segments to be restored first, bringing the application up sooner.

In addition to these functions based on application intelligence, hybrid cloud systems also have a process for prioritizing and reducing traditional block content as well. This prioritization isn’t just simple file tiering based on access history but is similar to the page weighting process used by internet search engines. It identifies the most critical data segments (another working set) and maintains performance locally, by moving these data to SSD. In a DR scenario, this working set represents the data which must be restored first in order to restart applications, again, reducing the effective recovery time.

Clones in the Cloud

With the combination of primary storage, an integrated cloud interface and writable snapshot technologies, these hybrid cloud systems can create a real-time copy of the primary data set in the cloud, easily. This process takes Microsoft VSS consistent snapshots of application-specific data volumes locally and copies them to the cloud. When changes are made, the cloud-based clone of that data volume is updated, drastically reducing the backup window in the process. Since these clones are stored in the cloud, an organization can retain as many as needed for multiple restore points, with no requirement to copy entire data sets off to another platform for archiving.

Recovery

When data is recovered, the on-site copy can be restored with the appropriate clone from the cloud, again, leveraging the knowledge of application-based working data sets and block tiering to reduce the amount of data transferred and minimize the recovery time. In another DR recovery scenario, the hybrid cloud storage systems themselves may be hosted by the cloud provider, enabling near-instant return to operation. In the most severe disaster scenario, where the primary data center is destroyed, users can upload a configuration file from a secondary location and have a new hybrid cloud system populated with data, starting with the most critical working set for each application.

Moving very large amounts of data off-site, and back, with relatively affordable replication software has not met expectations. Even the paradigm shift represented by cloud storage hasn’t generally lead to effective or cost-effective DR solutions for many companies. The reason is that an old ‘brute force’ approach to backup and DR, that was designed for local networks, has been reused on this new architecture of the cloud. This approach simply moved too much data, unnecessarily.

What’s needed is a new approach that works ‘smarter’ instead of just working harder, a new approach that uses intelligence about applications, data objects and their importance in time to identify the most critical ‘working sets’ of data. Then these can be kept on the highest available storage, be updated most often in the cloud and restored first to provide a dramatic improvement on backup and DR.

Hybrid cloud storage systems offer an alternative that can bring the advantages of cloud storage to bear and provide an economical, effective DR option. Using an intelligence about application-specific data and an automated mechanism for prioritizing data blocks, these storage systems can dramatically reduce data sets that need to be copied off-site - and reduce the time required to restore them as well. Leveraging available cloud provider services, these systems eliminate the need for dedicated off-site infrastructure and provide an answer for companies that need effective and affordable disaster recovery.

Eric Slack, Senior Analyst