Many IT pros see scalability as one of the key advantages of hyper-converged infrastructure -- but not all systems tout that benefit. Some are too heavily engineered, which makes it impossible to decouple the three constituents of a hyper-converged platform -- server, storage and network -- to grow one without growing the others.
Scale hyper-converged systems with servers
First, let's consider the three means of expansion for hyper-converged infrastructure when it comes to servers:
- Scale out: IT teams add new resources alongside existing ones through the implementation of commodity components. For example, your hyper-converged system may be a solidly engineered box with no direct expansion capabilities. If you run out of resources, you must buy another box and run it alongside the existing one -- or provision any old servers and just use data center connectivity to enable load balancing, data flows and so on.
- Scale up: IT teams add new resources through dedicated, engineered proprietary systems. For example, your hyper-converged system comes with a chassis that allows you to add more server systems when you need them. You'll need to buy a new motherboard, new ASICs, new PCI busses and so on.
- Scale through: IT teams deploy commodity resources into a hyper-converged system, and optimize them in a way that ensures the intelligence of the hyper-converged system can use them. For example, a hyper-converged system comes with the capability to add extra Xeon CPUs. When you run out of server capacity, you can just plug in an extra CPU to avoid the overhead of a new subsystem. The hyper-converged system embraces that extra resource and uses its own intelligence to make the most of it.
All three approaches above should cover the need for not just extra CPU capacity, but also for data paths and so on.
Scale hyper-converged with storage
These same expansion methods apply to hyper-converged storage. Some basic hyper-converged systems have a fixed storage capacity. When you reach that storage capacity, you must buy a complete extra hyper-converged system and plug it in alongside the existing system just to get that extra storage. While it is possible to plug in a network-attached storage or storage-array network system, you'll lose the advantages of a high-performance hyper-converged system.
Other systems will allow you to add extra storage through complete subsystems. However, just as the addition of a whole server subsystem replicates a lot of unnecessary capacity, this approach to storage expansion is also wasteful.
Instead, look for systems that do not require you to add storage controllers just to gain raw storage volume. As with CPU resources, look for storage subsystems that can accept extra drives such as solid-state drives or spinning disks, or that allow you to flexibly replace existing drives with higher-density drives. This is not a hard and fast rule, but you'll need the capability to do both, depending on your needs. For example, you may have enough storage volume, but not enough IOPS throughput. If you add extra drives, it will just worsen the issue if that IOPS constraint is down to the controller. Instead, add an extra controller and divide the drives up between the systems to better distribute IOPS across the multiple controllers.
Scale hyper-converged with networking
Some hyper-converged systems only support a given number of ports. Once you use up those ports, you must deploy and connect an externally connected network switch to the hyper-converged system via either low-cost 10 GbE, or via dedicated, higher-priced 40 GbE interconnects. A more robust hyper-converged system will instead allow for network expansion within the system itself through the addition of a direct set of extra ports. As with servers and storage, this may be a set of unintelligent ports that the hyper-converged switch capabilities manage directly, or it may be a separately managed, intelligent set of ports.
There is no single answer to whether a hyper-converged system should be scale out, scale up or scale through. Try to choose a hyper-converged platform that allows all three options, and then understand the workloads that run on that platform to best meet their needs.
IT scalability a plus of hyper-converged
What are the positives and negatives of a move to hyper-convergence
Use these apps to make the most of HCI scalability