BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
Data center managers face a steady stream of new demands on existing compute, storage and networking resources, while aiming to minimize costs and administrative overhead. This is no small feat.
Web companies, such as Google, Facebook and Amazon, have identified ways to create and sustain large-scale compute, storage and network services. Some of the same characteristics of Web-scale system design are influencing hyper-converged infrastructure (HCI).
Hyper-converged infrastructure is a hardware and software platform that integrates computer, storage and networking resources in commodity hardware. It is closely related to converged infrastructure with integrated hypervisor virtualization, and is touted as a way to avoid the complexities involved in IT scalability within a customized IT environment.
An HCI architecture tightly integrates storage with compute systems rather than relying on storage area networking or network-attached storage. Hyper-converged infrastructure is distributed and horizontally scalable; additional nodes are added to a cluster as resource requirements grow so that existing hardware need not be replaced with larger servers. HCI is typically based on the x86 processor architecture.
Combining compute, storage and network hardware in a single rack unit is useful, but virtualizing those resources generates more value. An HCI product puts the idea of software-defined infrastructure into use. Processors and storage are virtualized and allocated according to the needs of specific operations. This helps improve overall utilization of resources. While server virtualization has helped optimize the use of compute resources for decades, storage optimization has not always been easy.
It is relatively easy today to create virtual servers. Working with logical unit numbers, storage volumes and other lower-level storage components and abstractions, however, can slow deployment of a virtual server. The software designed for managing hyper-converged infrastructure integrates server and storage virtualization management. Plus, it reduces the administrative overhead typical of less integrated infrastructure approaches.
The Capex case
A hyper-converged infrastructure offers several notable benefits. For large organizations, this approach simplifies capital expenditure (Capex) planning for scalable IT.
Although Capex might still require a planning horizon of two to three years, it is a simple matter to incrementally add more capacity to a hyper-converged environment. There is no need to purchase for peak expected capacity initially if that peak demand is not anticipated for a year or more. This is a key advantage of horizontal IT scalability over vertical scaling-up, wherein IT administrators increase the capacity of the existing node or swap it out for a bigger, more powerful server.
Smaller organizations, too, can benefit from HCI. Storage management overhead will be lower. Companies that adopt a software-defined approach can benefit from DevOps, the management practice that focuses on automating infrastructure management and supporting continuous delivery of software improvements to applications.
The IT scalability payoff
Review the applications that will run on hyper-converged infrastructure to determine how well they function in a distributed environment.
Hyper-converged infrastructures are well suited to big data and analytic applications. Big data analytic platforms, such as Apache's Spark and Hadoop MapReduce, benefit from keeping data near compute resources and minimizing data transfers over the network.
HCI provides a particular boost in IT scalability to applications designed for distributed systems. NoSQL databases, such as Basho Technologies' Riak, MongoDB offerings and Apache's Cassandra, scale up as new servers are added to a database cluster on hyper-converged infrastructures. Relational databases scale horizontally to some degree, but the relative costs are typically higher and the benefit lower than for distributed systems.
Diversity in hyper-converged infrastructure
VMware has introduced EVO:RAIL, a hyper-converged appliance based on its vSphere virtualization technology. The appliances are delivered and supported by VMware partners. SimpliVity is known for its hardware-based deduplication and compression. Gridstore, meanwhile, focuses on Hyper-V hypervisor platforms. It addresses the need to customize compute and storage by offering flash in both compute and storage nodes. Nutanix is supporting the increasingly important area of hybrid cloud computing. As enterprises balance the benefits of on-premises computing and public infrastructure as a service, they will adopt platforms that support management of hybrid infrastructures, such as those offered by Nutanix.
Inefficiency and performance hits
There are potential drawbacks to hyper-converged architectures.
As noted, not all enterprise software architectures take full advantage of the IT scalability in a distributed platform.
A one-size-fits-all approach to designing nodes also leads to inefficiencies in some scenarios. Some applications require more compute resources than storage, while others are more storage-intensive. Look for vendors and products that allow the data center management team to find the right balance of compute and storage in a single node.
Since many hyper-converged infrastructure product lines are based on commodity processors and storage, the management software is a key differentiator. System administrators should consider how the software platform supports deployment and monitoring, as well as backup and recovery.
The compression and deduplication that hyper-converged infrastructures bring to IT deployments can improve overall storage efficiency, but at the risk of some loss in performance. Hardware-accelerated compression and deduplication can mitigate that risk.
Hyper-converged infrastructures significantly reduce costs and lower administrative overhead in applications that thrive on horizontal scalability. To optimize the return on investment, though, you may want to adopt DevOps practices that automate management tasks and support agile software development and deployment. No matter how much the hardware and software improves, it'll be management practices that really drive results.
Make a strong start on DevOps
Vendor choices for hyper-converged infrastructure
How HCI products use solid-state flash