The business value of replication is time. What is a minute of online orders worth to your company? How about an hour? How about 2 days? Longer outages can be fatal to your company if data is not available. Your customers start looking elsewhere, and the damage may not be reversible. The American financial system did not collapse on 9/11 because critical data was replicated in nearby New Jersey.
Replication is insurance against the unknown. There are different kinds of replication, and let’s look at three categories of potential outage: local, metro or regional. Local would include problems inside the data center, or in a company campus. LUN mirroring provides a current copy of critical data locally, as in the next rack, room or building. There is a consequence, however. Data must be written to both arrays before the application can move on to the next task. This is to assure the data is current in both arrays. This means that there is a small additional latency for this data bounded by the amount of time to move data between the two arrays.
For a metro area outage example, in the case of a building fire or some other disaster affecting the entire campus, data can be protected by using synchronous replication. Critical data can be synchronously replicated across the metro area to another data center. This means the data on both sides of the metro area are current and no data is lost if one site becomes unavailable for any reason. Since system latency will depend on the distance between the two sites, most synchronous sites are within 100km or 66 miles to keep the latency modest, and still provide the insurance against catastrophe. Hence there is a more latency to deal with than the LUN mirroring example.
An even better way to preserve uninterrupted access is to employ a stretch cluster so the application server is part of the architecture. In this way operations continue without interruption even if one of the sites becomes unusable. It is a way to make sure not just the data, but the applications are always available if there is a problem with the primary site.
100km may not be enough distance to protect against other kinds of calamity, however. In the case of an earthquake or typhoon/hurricane entire regions can be affected. This means a different technique should be used; asynchronous replication. Asynch replication does not require the data to be written to the remote site before the application goes to the next task. It does mean data can be lost. This is where journaling and time-stamps make recovery work. If the primary data site is under water, then the system in the recovery location with the asynch replication data will have a gap in data during the time the information was written in the primary site before it could be written at the remote site. Because these links to remote sites can be expensive, WAN optimization is important. WAN optimization includes data reduction to reduce the amount of data transmitted, encryption to keep it secure, and aggregation and failover capabilities to make the most of the effective bandwidth.
The benefit of asynchronous replication is that operations can continue from a known point in time, and business can continue. Continuous Data Protection provides granularity to restore your system to a specific point where known data was written, greatly easing recovery. Automated disaster recovery services also greatly ease the recovery effort, and get operations back quickly to reduce impact on the business.
You can combine these technologies to meet a variety of objectives. You might choose more than one technique for the same data, both synchronously and asynchronously replicating the same data in different places. To save on money, you might replicate to a deduplicated LUN, and keep the primary storage on a thick LUN to preserve maximum performance in the primary site and still benefit from lower costs in the recovery data center.
With Violin, you have the whole portfolio of data insurance solutions to potential problems: local, metro or regional. You mix and match as you need to achieve your availability and performance objectives, as well as cost constraints. The alternative to a comprehensive recovery plan is potentially going out of business. That’s why you might consider Violin Memory as data insurance for your business. If your storage vendor doesn’t have all these capabilities on an all-flash array, you don’t need to settle for a partial solution. Give us a call and see what we can do for you.