Perhaps the biggest stumbling block to deploying a Software Defined Data Center (SDCC) is the architecting process -- qualifying and validating every component to ensure that server, networking, and storage hardware will interoperate in a predictable manner with mutually compatible APIs. To put it another way, if two connected systems attempt to communicate using mutually unintelligible protocols, nothing will be accomplished, and the task of trying to integrate these pieces becomes substantially more laborious.
Many hardware vendors provide ready-to-deploy systems tied together with SDDC management software. This article highlights hardware vendors worth checking out before deploying an SDDC and some of their solutions that work for given use cases. What follows is far from exhaustive: other vendors may provide a more appropriate solution, or your organization may wish to work with a familiar vendor. In addition to the solutions highlighted here, SDDC solutions are also available from VCE, Nimble Storage, Hitachi Data Systems, IBM, Dell, Fujitsu, and Hewlett Packard Enterprise, among others.
NetApp's FlexPod line of 'converged infrastructure solutions' is designed to work with Cisco's Unified Computing System (UCS) servers and Nexus fabric switches to create a validated and reliable operational model for a variety of different workloads. NetApp touts the all-flash architecture of FlexPod as being able "to deliver up to 4 million IOPS with 1ms latencies." The big selling point of this is for database use, for which the company highlights compatibility with Microsoft SQL Server and Oracle.
To that end, the HammerDB benchmarking comparison performed by NetApp compared a storage array with 144 450GB 15K SAS platter hard drives, for which load was increased such that a consistent read latency of 20ms was experienced. An identical HammerDB configuration was subsequently deployed on an AFF8080 EX, which houses 48 400GB SSDs. NetApp reported a "20x reduction in I/O latencies, 4x improvement in storage IOPS, and 4x improvement in SQL Server CPU utilization," while claiming that the AFF8080 EX has sufficient headroom to deliver higher performance than what was shown in the benchmark comparison.
Cisco's UCS platform servers are available in various configurations. The B series is modeled for powered chassis or half-slot blade servers, while the C series is modeled for standard 19-inch racks, while UCS Mini systems are available for remote office and point-of-sale environments. The Cisco UCS Manager software allows for remote management access using an HTML5 web interface, a command line, or programmatically using an API, while UCS Central extends these functions across multiple domains, and optionally across physical locations.
Quanta Cloud Technology's QxStack is effectively an SDDC in a box -- the hardware is pre-qualified and pre-validated, and simply needs to be connected to power and the network. Deployment is greatly eased compared to custom-built solutions, as the system comes pre-imaged with information about the hardware shipped in the rack; it uses this information, plus user-provided environmental data, to initialize the rack.
The QCT QxStack consists of a minimum of eight QuantaGrid D51B-1U servers, powered by Intel Xeon E5-2600-class processors. The system can scale incrementally to up to 24 servers per rack, up to a total of eight racks, for a total maximum of 192 servers. The spine switches located on the first rack are two QuantaMesh BMS T5032-LY6 units, with each rack holding two BMS T3048-LY8 top-of-rack switches, and a dedicated BMS T1048-LB9 switch for management.
The automated hardware management of VMware's EVO SDDC is capable of automatically discovering new servers added to the system and adds those resources to the SDDC Manager for immediate use. Additionally, the lifecycle management system catalogues the hardware and software that comprises the SDDC, including version and patch information. The SDDC Manager is capable of processing system updates -- from low-level components such as firmware, BIOS, or drivers, or higher-level components such as VMware software bundles -- on a scheduled cadence, or as needed.
Fujitsu and Red Hat offer the PRIMEFLEX reference architecture for OpenStack that combines Fujitsu's server and storage solutions with Red Hat's branded distribution of OpenStack on top of Red Hat Enterprise Linux, as well as that company's support and service offerings. While OpenStack does not provide a hypervisor (in contrast to VMware SDDC), the role of hypervisor can easily be filled for an SDDC deployment with KVM, or with Docker if you prefer containers for your environment.
Fujitsu's PRIMERGY servers are qualified for use with Red Hat OpenStack, for which the company claims up to 27 percent reduction in cooling costs as a result of cool-safe advanced thermal design. Mass storage is handled by Fujitsu ETERNUS DX systems for traditional storage deployments, or optionally by the ETERNUS CD10000 S2, which uses the Ceph distributed storage platform. In addition to addressing a litany of issues in volume rebuilding inherent to the ageing RAID design, Ceph frees storage from the restrictive vendor lock-in of qualified hard disks and RAID controllers with custom, closed-source firmware.