Composable Infrastructure Evaluation Guide – Evaluator Group

This report The Composable Infrastructure Evaluation Guide was published on July 9, 2019, by Evaluator Group, Inc.

Evaluation Guide Overview
The guide is part of a series of evaluation guides designed to help evaluate technology alternatives. These guides are for IT professionals seeking a neutral, objective discussion of the design considerations behind new products, technologies and trends. They are not vendor-sponsored. They are developed based on reviews of the technology options and strategic analyses of how they can best be used in today’s information storage environments.

This guide focuses on technology education, explaining what composable means, where this technology came from and what to expect as it matures. It should be noted that this is a look at an early-stage technology and as such, available products, approaches and even defining characteristics are somewhat fluid. Consequently, this guide does not include EvaluScale, the comparison tool that ranks characteristics of different products in a given technology category.

Composable Infrastructure Definition
Evaluator Group defines composable infrastructure as a comprehensive, rack-scale compute environment that uses software-defined networking techniques to connect independent servers, storage and switch chassis or modules by PCIe or Ethernet. Designed for maximum efficiency and agility, it maps pooled, physical resources to create (or compose) bare-metal compute instances as needed, with a minimum of time and effort, often under programmatic control. It provides a dynamic compute environment for applications running on OSs, VMs or containers that also supports rapid reconfiguration, or dissolution back to the component resources. Composable systems are deployed and maintained by core IT personnel.

How We Got Here
There is an ongoing movement in IT to improve resource utilization, increase agility and control operational overhead, using the public cloud as a model, to some extent. IT also needs to keep up with the continuous and accelerating pace of applications development and deployment required by the modern enterprise. Mobile apps, the IoT and the rise of DevOps processes are all adding to the rate of change in IT infrastructure. IT must simply move faster. Integrated systems like converged and hyperconverged infrastructures are a response to this need accommodate accelerating change.

Server virtualization was a big step as it disassociated the compute process from hardware by defining the server itself in software. Converged Infrastructures (CIs) provided a faster route to deployment by bundling servers, storage systems and networking components, usually with a hypervisor, and selling them as a turnkey, rack-level compute environment.

Hyperconverged Infrastructures (HCIs) clustered multiple server nodes via Ethernet, typically running SDS as a VM on each, to pool physical storage devices and provide a common management interface that simplified deployment and operation. They provided more granular scaling, better agility and simple operation, but could not support the capacities that some large enterprises needed. They also were not particularly efficient at that scale as they added a software layer for storage virtualization (SDS) and a hypervisor (for most products).

Composable infrastructure borrows from these integrated systems technologies to create a more flexible and efficient infrastructure for large enterprises and service providers. As workloads become more dynamic, infrastructures must become more dynamic as well. To this end, the composable architecture is designed for API-based software control.

Composable Infrastructure Concepts
Composable seeks to reduce the time and effort required to deploy and configure compute infrastructure to support dynamic workloads and to improve the resource efficiency in large-scale environments. In this context, efficiency means to maximize utilization and deliver the highest performance potential to the application.

Composable infrastructure incorporates elements of both CI and HCI, disaggregating physical resources as CI does; putting storage, networking and compute into dedicated modules and not combining them as HCI does. This allows it to create compute configurations with more storage, more networking connectivity and special resources like GPUs and FPGAs.

This technology uses software, a composing engine, to control resource allocation. But, instead of virtualizing compute and storage with a hypervisor and SDS, it dynamically maps direct-connected storage, networking and other devices, like GPUs, to servers using virtual networking techniques.

Composable does not use SDS, but supports direct-attachment of storage devices (like drives and SSDs), eliminating the SDS software latency. It also does not require a hypervisor to run an SDS layer or workloads, but instead, creates bare-metal server instances that can support containers or a hypervisor if desired, reducing software licensing.

Composable infrastructures abstract the physical connection of resources to a server chassis using software-defined networking. They compose each compute instance, at run time, as if an administrator had cabled the needed storage devices, NICs and other components to the server for a specific application. Then, these resources are disconnected and returned to a common pool, in order to be composed again, into another configuration. Composable infrastructures can be controlled by management platforms through APIs, making this technology well suited for automation.

Stateless architecture
Although somewhat ambiguous, the term stateless is often used by composable vendors to describe their technologies. Most define this term as similar to persistence, where stateful compute instances save data and configuration information between compute sessions. In contrast, composable infrastructures essentially create each compute instance anew when the composing software connects storage and other resources to a CPU and memory with the appropriate boot information.

Characteristics
Evaluator Group has found composable infrastructures to have the following characteristics, although implementation of these may vary somewhat by offering. The following chart is designed to compare converged and hyperconverged infrastructures in terms of these characteristics.

Comparison of Composable Infrastructure Characteristics With CI and HCI