Although hyper-converged systems are often used for specific applications, such as virtual desktops, people regularly...
ask whether hyper-convergence can scale to meet the needs of an entire enterprise. As is the case with so many other things in the world of IT, the answer isn't cut and dry.
One of the main reasons for this is that hyper-converged products are not created equally. Each vendor chooses the hardware and software to be used in its hyper-converged systems, and these choices have a direct impact on the degree of scalability that can be achieved. Another factor that plays into the question of whether hyper-converged products are appropriate for the enterprise is each enterprise having unique needs.
Because each vendor's approach to hyper-convergence is different, most of them design their hyper-converged systems to be modular. Modules commonly take the form of nodes. A node is a self-contained resource that contains CPU, memory and network and storage resources.
When evaluating the suitability of hyper-converged products for enterprise use, you must remember that nodes can vary widely in terms of the resources they contain. A single vendor might offer a variety of node types as a way of appealing to different customers.
Nodes are often grouped together into appliances. An appliance is essentially a hardware chassis that can accommodate multiple nodes. For example, a single appliance might accommodate four nodes, each of which adheres to the 2U form factor. Appliances can impact scalability in the following ways:
- They consume floor space. As such, a data center only has room for so many appliances.
- Appliances are generally clustered together, so there is a software limit to the number of appliances that can exist within a single cluster. In the case of VMware's EVO:RAIL, for example, a single cluster can consist of up to 16 appliances. Each EVO:RAIL appliance can accommodate up to four nodes, so there is an effective limit of 64 nodes per cluster.
As previously explained, nodes can vary in terms of hardware capabilities. Given these differences and the number of nodes that can potentially be added to a cluster, clusters can vary significantly with regard to the workloads they can accommodate.
Hyper-converged products and software limits
While hardware capacity is of undeniable importance, there are also limitations built into the software. Hyper-converged systems use standard hypervisors such as Microsoft Hyper-V or VMware ESXi to host virtual machines (VMs).
Each hypervisor has a limit as to the number of VMs it can host, even if infinite hardware resources are available. There is usually a per-node limit and a per-cluster limit. EVO:RAIL has a per-cluster limit of 12,800 VMs, although the available hardware resources and workload requirements may reduce the actual number of VMs that can be hosted.
Of course, there is no rule stating that an organization is limited to using a single cluster. As is the case for non-hyper-converged clusters, tools such as VMware vCenter Server and Microsoft System Center Virtual Machine Manager can provide centralized management capabilities for multi-cluster deployments.
In terms of scalability, hyper-converged products tend to use the same hypervisors and management tools as non-hyper-converged systems. As such, hyper-converged systems can often be integrated into existing virtualized environments. The question of whether hyper-converged storage systems can scale to meet the demands of enterprise environments largely comes down to hardware.
Although there is nothing stopping an enterprise-class organization from adopting hyper-converged products, it is generally more cost effective to use fewer high-powered servers than to use numerous modestly equipped servers. Not only do the hardware costs tend to be lower, but using fewer servers may also decrease the administrative overhead. Granted, hyper-converged nodes tend to be more powerful than they once were, but these nodes are still limited by their physical size.
Another issue that may come into play is that of storage. Nodes in hyper-converged systems generally include their own built-in storage. Enterprise-class organizations prefer to use large, centrally located storage appliances rather than local storage. As such, the storage included in a hyper-converged node may increase costs without delivering any real benefits to an enterprise-class organization.
Hyper-converged systems could theoretically scale to the point that they would be of use in an enterprise environment. However, enterprise organizations may find it to be more cost effective to use non-hyper-converged resources, so they can spend money on needed resources and not those that may go unused.
Scale boosts hyper-converged products with flash
Options expand hyper-converged system offerings
A guide to recent hyper-converged product releases