Application performance drives business value, so it's important for an organization to extract the maximum performance...
for the money spent. Tuning can help get the best performance out of the hardware purchased.
In legacy architectures, compute and storage silos are fertile ground for performance tuning. However, hyper-converged platforms make much of the tuning unnecessary by replacing existing compute and storage silos with an integrated platform joined together by a single network. So where should administrators carry out performance tuning on a hyper-converged platform?
Above the hypervisor
The majority of performance tuning is done above the hypervisor, tuning the virtual machine configuration, operating system and applications inside the VM. Right-sizing virtual hardware so that the application inside the VM has adequate resources is important, as is optimizing the setup of the OS and the applications inside the VM. Good application and data design are the most effective optimizations, and a well-configured database application can perform far better than a poorly configured application. It is hard to make up for this scale of inefficiency anywhere but inside the badly written application. None of these things change on a hyper-converged platform.
Hypervisor and below
One of the objectives of hyper-convergence is to simplify the operation of the virtualization platform, allowing more time to focus on the applications inside the VMs. One aspect of this simplification is reducing or eliminating the need to do a lot of performance tuning of the hyper-converged platform. A fundamental design aim in every hyper-converged product is to remove the bottlenecks in traditional infrastructure. The next aim is to deliver ample resources so that contention between VMs is minimized. Network bottlenecks are removed by using 10 Gb Ethernet for VM traffic rather than multiple 1 Gb Ethernet switches that are each easier to saturate.
Storage bottlenecks are far more common on legacy architectures, overloaded storage network links or storage processors in the array. Hyper-converged platforms effectively put a storage controller inside each node, changing the storage network topology and providing far more storage controllers. The hyper-converged platform also builds in all the optimization of the hypervisor that is needed to work with the hyper-converged storage. The hypervisor is pre-tuned by the hyper-converged vendor.
Sizing is important
The primary performance tuning control in a hyper-converged platform is node sizing. Unfortunately, this is only done when new hardware is purchased. Once nodes are bought, adding RAM is usually the only upgrade option. For CPU and RAM, you generally want a node that is several times the size of the largest VM. This gives hypervisor load balancing the best flexibility in placing VMs across the nodes. The tricky part is storage performance. We want the frequently accessed blocks of the VM in the fastest storage tier, but we seldom know how much VM data is frequently accessed. The result is a rule of thumb: Use 10% fast flash for most workloads and all-flash for critical workloads.
Hyper-convergence is about simplifying IT infrastructure. Removing the need for performance tuning simplifies management. Most hyper-converged platforms don't have tuning knobs for the infrastructure. Tuning VMs, OSes and applications remains a constant part of IT operations.
Could your hypervisor cause slow VM performance?
Complications a hyper-converged platform may bring to your data center
Hyper-converged infrastructure aims to solve integration issues