refresh(PIX) - Fotolia
For many years, the traditional IT model has been to build out separate towers of technology, covering servers/virtualisation, storage and networking.
But hyper-converged infrastructure (HCI) has changed the landscape and the thinking around IT deployment.
Hyper-converged systems deliver all parts of a typical IT infrastructure out of a single set of hardware and software that performs the previously segregated tasks of storage, networking and compute.
In this article, we look at the options to deploy hyper-converged infrastructure via software products.
The advantage of moving to hyper-converged is capex- and opex-driven. HCI optimises the use of hardware and uses it to deliver compute – virtual machines (VMs) and containers – and storage functionality.
Modern processor and networking architecture means there is little or no need for dedicated or custom hardware, so even complex storage functionality can easily be delivered out of software.
At an operational level, hyper-converged infrastructure simplifies the deployment, ongoing management and refresh of hardware, while moving to a more cloud-like consumption model.
Additional capacity can usually be implemented by deploying extra hardware nodes and adding them to the existing configuration. The protection of data across the hardware is managed by software, providing graceful recovery from component or server failure.
The final piece of operational benefit comes from deploying certified and validated configurations.
Companies such as Nutanix, Simplivity (now HPE) and Scale Computing are dedicated hyper-converged suppliers, offering hardware that has been pre-tested and evolved to meet customer demands for performance and scalability.
But there are numerous ways to deploy hyper-converged from software products.
First, implement a resilient storage layer. If a storage device or node fails, recovery should be automated with the minimum impact on performance.
When nodes (servers) are added or removed, storage resources should be automatically re-balanced, preferably with asymmetric configurations (where nodes have different hardware specifications).
At the hypervisor layer, creation of VMs (and potentially containers) needs to be directly integrated into the orchestration platform.
Typically, the administrator should not be bothered with having to decide where VMs will be placed on storage. Ideally, any specifications should talk about performance (for compute or storage) in abstract terms based on policy and service levels, rather than as functions of the hardware.
Networking at the physical level is usually implemented through dedicated hardware and configured in software as part of the hypervisor. In this way, networking switches act as a backbone for traffic between nodes and to the outside world.
So what choices are available in the market today if HCI is the preferred deployment model for infrastructure?
One approach is to work with the companies, such as those mentioned above, that offer off-the-shelf hyper-converged products.
There is also the option of looking at a completely customer-designed and implemented system and only use software from the supplier.
Finally, there are reference architectures that specify recommended and tested components, but are not sold under a single stock keeping unit (SKU).
Software-only options allow customers to fully customise their hardware options, with minimum specifications or recommendations from the software supplier.
The software-only option allows the use of existing hardware resources or a preferred supplier, which enables hyper-converged to be encompassed within current hardware agreements and support models.
IT departments that go down the software-only route need to be happy developing and maintaining their own hardware compatibility lists, including firmware for components such as networking and storage.
Here are examples of software that can be used to build out DIY hyper-converged.
VMware provides the capability to build DIY hyper-converged around its Virtual SAN or vSAN product. vSAN was introduced in March 2014 as a kernel-integrated storage layer that turns multiple VMware ESXi hosts into a distributed storage platform. As a flagship product, vSAN has been rapidly developed and enhanced with new features, including erasure coding, encryption, all-flash support and stretched clusters.
Integrating vSAN with other existing VMware products such as vRealize Automation suite allows customers to build a complete hyper-converged system around the vSphere platform.
However, vSAN and vSphere products must conform to the VMware Hardware Compatibility Guide, which for vSAN covers specific HBAs and SSDs, so hardware choices are focused on those already tested and validated by VMware.
Virtual SAN is licensed by node (for vSphere servers either consuming or providing storage resources), with a minimum of four nodes recommended.
Read more on hyper-converged infrastructure
- Converged and hyper-converged infrastructure are ready-tested ways to put compute and storage into the datacentre. But which should you choose for your deployment?
- The rise of hyper-converged infrastructure – with compute, storage and networks in one box – seems ideal for SMEs, but is it always a better idea than traditional IT architecture?
HPE allows customers to build hyper-converged systems based on StoreVirtual VSA software.
StoreVirtual is an appliance that runs as a VM guest across multiple hardware nodes and provides a scale-out storage infrastructure that can be integrated into either VMware vSphere or Microsoft System Center (with Hyper-V) virtual environments.
StoreVirtual implements a network-based RAID architecture that can easily be extended as additional servers/nodes are added to a hardware configuration. Following HPE’s recent acquisition of Simplivity, it remains to be seen whether StoreVirtual VSA will remain or be replaced by a software-only Simplivity option.
More interesting for HPE is the continued development of the Synergy platform that allows hardware components to be combined to make many logical configurations.
With Synergy server nodes, DAS storage and StoreVirtual VSA, customers can create hyper-converged systems without buying dedicated storage. These configurations can be reconfigured on-demand through software, providing many options to deliver infrastructure in highly dynamic environments.
Maxta is a storage startup that originally focused on providing a distributed storage platform called the Maxta Storage Platform (MxSP), but has moved to DIY and appliance-based HCI.
MxSP aggregates resources across multiple hypervisors to provide a single scale-out storage platform with multi-tenant support, clones, snapshots and data protection.
With MxSP, hyper-converged infrastructure can be built across platforms including VMware vSphere, KVM and OpenStack.
Maxta provides a single interface called MxInsight to manage heterogeneous environments, or integration into the hypervisor (as a vSphere web client or OpenStack Horizon plugin). Management is also supported through CLI and REST API.
Atlantis Computing offers the USX platform to build out hyper-converged systems.
USX is deployed as a VM across a cluster of hypervisor nodes that aggregate storage into a distributed pool of resources using DRAM and cache for read I/O acceleration.
Most recently, Atlantis Computing moved to focus on VDI and in July 2017, assets of the company were acquired by Hive-IO, so it remains to be seen whether USX will continue as a hyper-converged option for general compute.
StorMagic SvSAN is software for building out hyper-converged infrastructure that is focused on edge or small and medium-sized enterprise (SME)-sized configurations.
SvSAN has the benefit of being able to run with a minimum of two nodes, making it more cost-effective for organisations that need to support many smaller deployments, such as retail offices.
Currently, SvSAN supports Microsoft Hyper-V and VMware vSphere deployments, with a relatively light overhead of only one vCPU core and 1GB of DRAM.
StarWind Virtual SAN now supports the ability to build out hyper-converged infrastructure across Hyper-V, vSphere, KVM and XenServer environments, either with native integration (Windows) or running as a VM.
Virtual SAN uses the features of the Windows platform (most recently integrating with Microsoft Windows Storage Spaces) to provide scale-out storage across multiple Windows servers or virtual machines.
StarWind Virtual SAN software is also available in a free version that allows customers to evaluate before considering a purchase.
Finally, there are also systems based on OpenStack, such as ZeroStack and Stratoscale.
ZeroStack, for example, separates orchestration and management out as a software-as-a-service (SaaS) offering, while retaining physical resources on-premise.
This may point to the future, where we focus less on the hardware and more on the functionality, with management and orchestration offered across a range of platforms at the same time, including a mix of public and private cloud.