They have realised that not only is storage virtualisation and abstraction a crucial part of the hyper-convergence story, but it also provides a good template for abstracting the other layers of the system.
The big names have responded to the upstarts in kind, however. They might be best known as hardware suppliers, but they are no slouches at providing appliances and other packaged bundles. It helps too that, just like hardware, some of the necessary software is approaching commodity status.
The big names also claim they can do a better job of integration than the software-defined suppliers. Some offer scaling that is more flexible than a one-size-fits-all appliance approach, for example.
As well as supplying hardware for some software-based appliances, Cisco has its own offering for VMware-based convergence, which it calls the HyperFlex HX Data Platform.
A HyperFlex cluster features three or more HX-Series hardware nodes, each with two CPUs, up to 768GB of RAM, 10 Gbps Ethernet, a mix of flash storage and hard disk, and a controller that co-operatively implements a distributed file system across the cluster.
The latter uses an SSD layer for caching and hard disk for capacity. Replication to other nodes means that even cache writes can be acknowledged as persistent. As is increasingly common with such file systems, inline data deduplication and compression is applied across the storage layers and the cluster.
As one might expect from Cisco, HyperFlex also majors on its network capability, including the ability to automate the provisioning and configuration of network resources.
Its Dell XC nodes are built on PowerEdge servers and offer up to 20 cores and 1.5TB RAM per appliance. Storage is typically hybrid, with nodes that support flash and spinning disk hard drives, but there is also an all-flash model available with 10 800GB or 1.6TB flash drives.
Nutanix’s own boxes can have up to four server nodes per appliance, whereas Dell’s appliances are single-node, albeit with dual processors.
Another difference is that Dell offers storage capacity nodes, which Nutanix does not. These are only available for the KVM-based Acropolis Hypervisor, but are useful for expanding storage-constrained clusters, as they avoid the cost of buying unnecessary extra compute capacity.
PrimeFlex is Fujitsu’s range of pre-configured and ready-to-run packages for specific datacentre workloads. While the list includes typical hyper-converged platforms such as VMware and OpenStack, there are also PrimeFlex integrated packages for the likes of Hadoop, SAP Hana and Microsoft Sharepoint.
For example, its PrimeFlex cluster-in-a-box is a relatively small hyper-converged appliance that adds two or four server nodes to a Fujitsu Primergy CX400 shared storage device. Aimed at smaller organisations and branches, it runs Windows Server and sets up as a virtualised high-availability cluster.
HP Enterprise ConvergedSystem
A purpose-built appliance for virtualised environments, Hewlett Packard Enterprise’s (HPE’s) ConvergedSystem 250-HC is built on its StoreVirtual Virtual SAN Appliance (VSA) software.
This provides scale-out storage for the hyper-converged appliances that make up a ConvergedSystem cluster. A cluster can scale from three to eight appliances for a maximum of 32 nodes. Each appliance can host from two to four server nodes, giving it up to 96 processor cores and up to 2TB of memory, plus a mix of flash drives and HDDs.
The CS 250 cluster runs vSphere, with the addition of HPE OneView InstantOn management software to look after deployment and provisioning. Based on VSA, the cluster can also be expanded with storage-only nodes in a range of configurations, or connected to an existing VSA setup.
Off-site replication is included too, via StoreVirtual’s remote copy features, as is integrated connectivity with Microsoft Azure for cloud backup and disaster recovery.
Last but by no means least, VMware certainly has a lot of hardware partners working with its hyper-convergence offering EVO:RAIL, such as Dell, EMC, Hitachi, HPE and NetApp.
On the one hand, this is great for buyers. Choose EVO:Rail and you get the comfort of a big brand plus your choice of several hardware suppliers.
On the other hand, VMware seeks to minimise its support overheads, so it applies relatively tight hardware configuration requirements that leave the hardware companies less room for differentiation.
Interestingly, most of VMware’s EVO:Rail hardware partners also offer alternative hyper-converged packages. They typically suggest that this is to meet the needs of customers seeking different levels of scalability.
However, cynics have claimed the real reason is that they signed up to EVO:Rail merely to keep their existing relationships with VMware on track, and that they would rather sell their own hyper-convergence packages.
Either way, by incorporating core VMware tools, such as vCenter Server and Virtual SAN, EVO:Rail is designed to appeal both to experienced VMware admins and to new users seeking something easy to run.
Its hardware specification calls for the ability to scale out to four 2U converged appliances. Each of these appliances would typically run dual multi-core Intel processors and support 192GB of memory, plus 10 Gbps Ethernet and up to 4TB of spinning disk or SSD storage.
Read more about hyper-converged infrastructure
- In the first of a two-part survey, we look at the hyper-converged infrastructure market and the startups providing VM-native servers and storage, and datacentre-in-a-box products.