Converged infrastructure and hyper-converged infrastructure both offer methods to manage compute and storage resources...
more efficiently than traditional approaches. However, each has its limitations.
Startup Datrium takes another tack, which it calls open converged infrastructure (OCI). Datrium claims it can address many of the limitations of converged infrastructure (CI) and hyper-converged infrastructure (HCI) while still providing many of the same benefits as CI and HCI by offering a more flexible platform that completely separates the compute and storage layers.
In 2016, a relatively young company named Datrium introduced a different type of converged system based on the principles of open convergence. Datrium's DVX product takes HCI to the next level by offering OCI that separates the compute resources from the storage resources, making it possible to optimize and scale them separately.
As with HCI, the OCI platform uses virtualization and software-driven technologies to simplify deployments and management. However, the OCI model provides more flexibility than either CI or HCI. Datrium's open converged infrastructure DVX platform uses three primary building blocks: compute nodes, data nodes and the Datrium software that ties them together.
Compute nodes in an open converged infrastructure
The compute nodes do all the heavy lifting. Not only does each node come equipped with the processing and networking resources necessary to drive workloads, they also come with internal flash storage that serves as a local read cache to maximize I/O operations.
Workloads are normally deployed within virtual machines (VMs) based on one of the supported hypervisors -- VMware vSphere, Red Hat Virtualization or CentOS. However, Datrium also supports stateful Linux containers and Docker persistent volumes, both of which can be deployed to bare-metal or virtualized environments.
When setting up OCI environments, administrators can use Datrium servers for the compute nodes or they can use their own servers -- as long as they meet the minimum requirements to support the DVX platform.
The compute nodes have no dependencies on each other except when VMware vMotion is migrating VMs between systems. As a result, the servers can be configured independently to support varying workloads.
For example, administrators can implement beefier servers to host VMs that require greater performance and use lower-end servers to handle less-critical workloads all within the same OCI environment. Administrators can also use different hypervisors on the servers or mix virtual and container deployments to provide even more flexibility for how workloads are distributed and managed.
The compute nodes also run independently of the data nodes, which store all the data and deduplicate it at a global level. The data nodes are made up of fully redundant, hot-swappable components. The disks themselves can be spinning hard disks or Datrium's Flash End-to-End drives.
Integral to the DVX open converged infrastructure platform is the Datrium software that runs on the compute nodes. The software leverages local CPU resources to provide data services. Administrators can configure the software to use up to 40% of the server's resources. The software manages the data, controls I/O operations and delivers data services such as encryption, compression, data reduction and erasure coding.
The compute nodes communicate with the data nodes over dedicated network links. Data is written synchronously to nonvolatile random access memory on the data nodes to support faster writes. The data stored on the data nodes is presented to the compute nodes as a single network file system (NFS).
The Datrium software makes it possible to maintain the compute nodes and data nodes as separate entities while still providing a unified system. This also means that administrators can scale the compute nodes separately from the data nodes. Currently, Datrium's DVX platform supports up to 128 compute nodes and up to 10 data nodes, making it possible to provide as much as 1 petabyte of usable storage capacity.
Although Datrium is getting the most press right now with its OCI platforms, other companies are also venturing into OCI territory. For example, Hitachi Vantara has added open converged systems to its portfolio. In addition, Egenera -- which began using the term open converged infrastructure back in 2010 -- is providing OCI systems that run its PAN Manager software on either the IBM Flex System or BladeCenter systems. And IT Blueprint is offering OCI products alongside its HCI systems.
With technology partners such as Amazon, Cohesity, Docker, Citrix, Oracle, VMware and Microsoft, Datrium is well on its way to making a name for itself. In fact, the DVX platform has already found its way into such industries as healthcare, government, manufacturing, entertainment and financial services.
How OCI differs from CI and HCI
CI combines compute, storage, network and virtualization components into an integrated platform optimized for specific workloads. The platform virtualizes end-to-end resources and presents them with a single resource pool.
The storage in an OCI is directly attached to the other components within a rack structure to create a virtual storage area network that's integrated with the server and network resources. Each component in a CI implementation is highly optimized and engineered to provide a unified system.
Unlike a CI system, which takes a hardware-based approach to delivering services, HCI platforms incorporate a software-based architecture that integrates compute, storage and network resources into a unified platform. The infrastructure is made up of multiple self-contained nodes, with each node providing both compute and storage resources.
As with CI, the HCI approach uses virtualization to deliver a single resource pool. However, HCI's software-driven architecture addresses some of CI's limitations that result from interfacing directly with hardware components.
The open converged infrastructure platform offers advantages over both CI and HCI. Because OCI uses a software-driven architecture, it does not have CI's hardware-based limitations. This is especially important when it comes to abstracting the data layer from the workloads running on compute nodes, allowing the data to be presented as a single NFS regardless of where it resides in the data nodes.
Despite OCI's advantages, CI and HCI have the advantage of having been in the market longer than OCI and, as such, have been poked and prodded and put through the ringer in a wide range of environments. Open converged infrastructure has yet to be field-tested to the extent of CI and HCI, and it still has to prove itself before it can live up to the claims of companies like Datrium that it is the next generation of convergence.