HP leads in Gartner Magic Quadrant for Modular Servers (May 2016)

Modular Servers Overview

Gartner defines the modular server market as a class of server where the compute capabilities are delivered in a modular style and housed in a common, often proprietary, chassis. Storage and network components may also be housed within the same chassis housing. There is a broad range of modular servers that vary in configuration options, based on CPU type (reduced instruction set computer [RISC], x86 and system-on-chip [SoC]), memory and storage attachment, and switching technologies. Server design focuses on density of components and richness of infrastructure management software platforms. Modular servers have evolved from many years of blade, multinode and custom-made high-performance computing (HPC), supercomputing, and IT service provider designs. The term describes a style of server design that focuses on the ability to improve server management visibility, CPU architecture flexibility, server density and operational efficiency. Modular servers will be increasingly positioned to address many emerging web-scale IT and Mode 2 workload requirements.

The Modular Server Market

For Gartner to consider a server as “modular,” the device must fit into a chassis or enclosure that allows for the easy and rapid addition or replacement of new hardware. We do not consider rack, tower and frame-based servers as modular servers; we also exclude “build your own” servers created from custom-made motherboard and component acquisitions. Blade servers and multinode servers comprise the great majority of this market (see “Market Definitions and Methodology: Servers” for more information).

The design of modular servers focuses on the density of hardware components and the ease by which technology can be extended and/or replaced. Blade servers additionally offer the benefits of a highly integrated switched backplane that benefits input/output (I/O) performance and infrastructure management. All modular servers boast CPU architecture flexibility and support the requirements of many emerging, web-scale IT workloads. A wide range of modular server offerings is now available. These servers vary in configuration options, such as CPU type and memory, storage, and switching technologies.

Modular servers typically address smaller workloads and workloads that can scale out to achieve greater throughput. Blade servers are particularly well-suited for data center consolidation and modernization strategies, and most first-generation integrated system strategies have been based on blade servers. There are still some key differences between blade servers and multinode servers; blade servers can support scale-up as well as scale-out workloads (especially blades from vendors like Hitachi and Hewlett Packard Enterprise that support processor aggregation across multiple blades), whereas multinode servers only support scale-out workloads. The biggest difference is that a blade server will likely have a dedicated switch, while multinode servers are more than likely going to have some form of Ethernet-based switching. But the lines of separation between blades and multinode are blurring more and more with each generation.

The primary adopters of blade servers are large and midsize enterprises that desire maximum flexibility in their on-premises data centers. Most multinode demand is shared between large enterprises building out infrastructure for big data analytics, and external IT service providers (including those using hyperscale) that invest in massive scale-out architectures for public and hybrid cloud deployment.

So, the technology preferences for modular servers vary. Large enterprises turn to the technology for its sophisticated management capabilities, hardware density and ease of installation. External IT service providers, on the other hand, need only basic management capabilities and use the technology to reduce costs and increase agility, delivering IT services through lean, repeatable and dense system configurations. But all users have to ensure that the rack density of modular servers does not become a burden. A rack full of modular servers can exceed the heat dissipation capability of air-cooled racks, and power consumption has been a challenge for some power-constrained data centers as well. Data center leaders must always check that intense deployment of modular servers — especially very large-scale clusters — will not exceed the capability of their data center facilities.

The different use cases fit well with the concept of bimodal investment. In Mode 1 of bimodal IT, traditional IT teams need servers that are suited to business-critical workloads — such as blade servers — that justify investment in highly available hardware and services. However, in the Mode 2 scenario, a development team within infrastructure and operations will need servers where agility at the lowest cost is a leading characteristic — a situation better-suited to multinode servers. As a result, data center managers will have to decide whether to bifurcate their infrastructure sourcing strategy and create separate silos, or find a single solution that fits all their needs. From a workload perspective, a blade server design could address both sets of requirements. But it is likely to be overengineered and overpriced for the needs of most Mode 2 workloads.

The bifurcation of the industry creates new challenges — and opportunities — for vendors. In reality, IT infrastructure has always been bifurcated, where HPC investments have usually been driven by different teams within the enterprise — often reporting to the line of business rather than the IT organization. From an architectural perspective, the characteristics of an HPC workload are very similar to those of many new-generation cloud or analytics workloads. So Mode 2 is providing an opportunity for established HPC vendors like Atos, Cray and SGI to expand their coverage to include new market needs, and potentially create bridgeheads into the legacy data center that have not been possible before now. Meanwhile, vendors with proven Mode 1 track records are benefiting from the trends in HPC that increasingly allow multinode, high-volume servers to address HPC workloads that were once the reserve of more specialized hardware.

Gartner Magic Quadrant for Modular Servers 2016


Source: Gartner