Software-defined storage (SDS) abstracts storage services from the physical storage hardware and allows them to run on a dedicated appliance or as a virtual machine. It essentially replaced the features that were already there -- like LUN masking, snapshots, and replication -- and lets those features be commonly applied across a variety of storage hardware. But the layers of the datacenter -- the compute, the network, and the storage -- were all still very present.
Converged storage architectures
Converged storage begins the attempt to collapse these layers by providing dedicated hardware nodes that can perform compute, networking, and storage all on a single layer. The storage component of this single layer is driven by software and leverages the internal storage of each node to create a shared pool of storage that all the virtual machines in the cluster can access. This allows functions like virtual machine migration to continue to operate as they would in a traditional share environment.
[But does it have flash? Read Flash Storage: What's Your Best Option?]
The goal of converged storage architectures is to provide a turnkey and consolidated infrastructure that combines the three layers. Vendors essentially provide appliances to offer that turnkey experience. These appliances allow for rapid deployment but at the potential loss of flexibility. That is not necessarily bad, it just depends on the priorities of the organization. If speed of deployment is the most critical aspect of a project, then converged architectures can be very appealing.
Hyper-converged storage architectures
Hyper-converged architectures take the converged concept to the next level in that they are provided as software and they can run on any vendor's server hardware. This will appeal to organizations that have a long lasting relationship with a particular server vendor or to an organization that is looking to drive as much potential cost out of their storage infrastructure.
How converged is converged?
While converged architectures claim to collapse the three layers of the datacenter, they don't make them vanish. There is still plenty of networking and storage in these environments. You have to interconnect the nodes and you have to put storage into those nodes. The components that make up that storage or networking is another thing that differentiates the converged storage architecture from the hyper converged. In a converged architecture the components that make up these layers are selected for you. In hyper converged you get to or have to choose. While choice can be a good thing, you have to make sure you have the time to go through a selection and qualification process. All storage and networking hardware is not created equal.
There is also a group of storage offerings that claim to be converged. But in my opinion they are not. In these systems there are still three discrete layers; a compute layer where the hypervisor and VMs reside, a networking layer that communicates storage and messaging traffic, and a storage layer that houses the media and manages data services. The key is that the compute layer is not leveraged to support any of these other functions, while in both converged and hyper-converged it is.
There is nothing wrong with having three separate layers; we've done that for years and many datacenters manage these infrastructures quite well. But these are not converged architectures. At best they are pre-integrated solutions. As I stated above, even converged and hyper converged architectures still have networks and storage. In my next column we will talk about how to select the right components for those.
Could the growing movement toward open source hardware rewrite the rules for computer and networking hardware the way Linux, Apache, and Android have for software? Also in the Open Source Hardware issue of InformationWeek: Mark Hurd explains his "once-in-a-career opportunity" at Oracle.