It’s a floor wax and a dessert topping: Hybrid cloud storage

In a classic Saturday Night Live parody commercial Gilda Radner and Dan Aykroyd play a couple having an argument over whether New Shimmer is a floor wax or a dessert topping. Chevy Chase, as a product pitchman, shows them that "New Shimmer is a floor wax AND a dessert topping!"

Many may wonder whether hybrid storage clouds represent a similarly unholy pairing, combining all of the upfront investment and capital of an on-premise solution with the performance and availability concerns of the public cloud.

As is apparent from my past three posts, I believe that public cloud storage is a compelling solution for certain combinations of workloads, economics and organizational capacities, but not for all such combinations.

However, many organizations have a broad range of requirements. Is there a case for such an organization to adopt a hybrid strategy, distributing storage requirements across an on-premise storage pool and one or more public service providers?

At least three use cases come to mind...

Disaster recovery

Traditionally, establishing a viable disaster recovery plan was an expensive and often impractical endeavor. Organizations invested millions in setting up geographically-disparate operations centers complete with large amounts of idle computer equipment, underutilized power/cooling/etc., and very little conception of how they would actually activate the site or get administrators to the site in the event of a disaster.

By contrast, cloud service providers promise the ability to have multiple geographic facilities, which don't involve large upfront capital commitments, don't necessarily need to be idle (easier said about computing resources than storage resources), and which could be easily accessed and managed from anywhere the administrators happen to be.

Load balancing/variable load handling

For workloads where storage needs are unpredictable, bursty, or highly variable, public cloud storage may offer a more efficient and effective safety valve than adding additional on-premise storage. In this case, the (more expensive) cloud storage is created and torn down as needed.

All-cloud applications

These are use cases where both the applications and the data can run in the cloud, thus mitigating some of bandwidth and latency constraints of the Internet. For example, we have customers who have been quite successful migrating compute and data-bursty applications (e.g. GIS, rendering) to the cloud, while maintaining more conventional applications within their data centers.

In all three of the above cases, the public cloud storage is integrated with the on-premise storage, rather than simply an appendage. 

What is needed for such a model to work? Assuming that storage in multiple locations is to be treated as a unified, commoditized, virtualized and centrally managed pool, we need to have a complete stack that allows it to be treated as such. This includes management, virtualization and server O/S layers.

More specifically, the storage software needs to have many of the properties that I have discussed in earlier posts -- little box storage -- on steroids, if you will.

The storage software (or really, the storage operating system), must be capable of supporting a global namespace that scales out along the key dimensions discussed earlier (performance, capacity, and availability). It must be capable of: operating at petabyte scale; self healing and centralized management to cope with inevitable device-level failures; and, if the storage pool is to support both legacy and future applications, it should be broadly capable of supporting both file-based and object-based storage models, as well as providing a path forward for "big data" workloads. Finally, while it may be possible to migrate virtual machine images quickly across the Internet, it is not feasible to do so with terabytes of data. Thus, if multiple locations are to be available for on-demand compute and storage, the data must be seamlessly (and continuously) replicated, so that up-to-date versions of data are available in multiple locations in a timely manner.


Is this complicated? Certainly. Has industry delivered all of the necessary components of the stack? Not quite, but rapid progress is being made.

In the end, I believe that hybrid cloud storage represents a powerful and useful, albeit complicated solution. For now, the relevant SNL comparison is less like "New Shimmer" and more like another Gilda Radner SNL classic, "The Nerds."

Ben Golub was CEO of Gluster, Inc., which is now the Storage Business Unit of Red Hat. He is on Twitter @golubbe.  

Copyright © 2012 IDG Communications, Inc.

Shop Tech Products at Amazon