The Cloud As Primary Storage
January 19, 2010
To this point and for the foreseeable future, cloud storage will largely be isolated to use as an archive or as a backup destination. At some point, however, cloud providers are going to start offering cloud storage as a primary storage location. What needs to happen to get there and how close are we? This differs from the discussion about primary storage that a software as a service cloud (Saas) provider will use. As I wrote about in this blog last year, these solutions will require the typical tier 1 storage capabilities. This blog is about secondary tier applications that don't have the very high performance demands of the primary applications within a data center.
Obviously, for cloud storage to work as a primary storage vehicle, we are going to need to have very fast connections to relatively close storage pods. Zetta, as an example, is basically offering this configuration today. They are leveraging their own scale-out, clustered NAS configurations (common in cloud) and adding in a better focus on performance while developing local storage repositories.
Another option is to leverage the hybrid model of using a local caching appliance that is already developing in the archive and backup use of the cloud by companies like Iron Mountain, Axcient, Nirvanix and others. Why couldn't these caching appliances be upgraded to be a store for primary data as well? There is a data connection speed issue to overcome but fast connectivity, especially on a metropolitan level, can be had.
There is a data services issue. Clearly NAS vendors have invested many development hours in snapshots, replication, provisioning and a host of other expected NAS capabilities. Basic NAS services however could clearly be OEM'd and put onto these devices. Both NAS and cloud vendors would need to offer some form of a global file system that can transparently position files as they age. This would allow small, fast local storage that could spill over to a cloud storage repository. Of course, they would have to support a mixed backend of storage. A downside to many NAS global file systems is that vendors only support their own systems. Stand alone file virtualization like those from F5, EMC's Rainfinity and AutoVirt could have a role to play here.
NAS as primary storage in the cloud can be done, and sooner than I think many people realize...What about block storage? I'll discuss that next.