David Merrill

Hyper-converged Economics Part II : Non-correlated Growth

Blog Post created by David Merrill on Sep 8, 2016

Appliance-based hyper-converged (HC) systems have many advantages, including a building-block approach to CPU, memory and disk. The appliances are easy to purchase, highly integrated and simple to plug and play. The overall approach may seem to provide ease-of-use, and therefore lower total costs. We need to be careful to understand the TCO impact of HC as it relates to the VM size, workload, and growth factors. In my next (and final) blog on this topic, I will talk about the workloads and situations where HC is optimal. But in this entry I want to explain the issue of non-correlated growth and how this can hurt appliance-based HC TCO.


When a HC appliance is purchased, the CPU, memory and storage assets are fixed into the appliance. When you run out of one of the resources (disk for example), you purchase another appliance. The growth of HC using appliances introduces wasted resources, since the growth is dictated by one of the 3 resources. If you need more CPUs for the VM, then you purchase another appliance, thereby wasting the memory and storage. Since the 3 elements are fixed in the appliances, the growth applies uniformly to all 3, whether all the resources are needed or not.


The result of appliance-based growth tends to be an increase in wasted resources. Another way of stating the problem for the CFO would been a poor return on assets - or ROA. Nobody on the finance team wants to hear about poor ROA.


Wasted assets, such as storage, seems to be a minor thing, since disk is cheap. But the waste in physical assets triggers additional cost of waste in other areas:

  • Maintenance - customers pay for HW maintenance when the hardware is not used, or poorly used. Maintenance is a fixed rage for all elements, irrespective of their utilization factor
  • Power and cooling of under-utilized assets increases carbon emissions and the total power bill (kWatt per VM, or kWatt per TB)
  • Floor space
  • License fees, many software fees are based on sockets or core CPU. If these are purchased and wasted because extra storage was needed, the software license increases unnecessarily
  • Some appliance vendors employ a proprietary file system, so if you do run out of space you cannot simply mount a new volume of shared storage. You have to grow within the proprietary environment.
  • As the non-correlated growth of appliances increases (non-linearly to the VM growth) then eventually the management and labor costs will also increase


The alternative is simply to engineer hyper-converged systems without a fixed config appliance. These systems allow for storage growth, without purchasing additional CPU or memory (or licenses). Applications that have scale-out storage requirements are not a good fit for appliance-based HC. Similarly, if the workloads can change the requirements for CPU or memory, the non-appliance based approach allows for more blades to be added without the accompanying storage.


HDS offers a wide range of compute, storage solutions in a hyper-converged to converted. This family of offerings does not create new islands of assets, but a common management and scale-up or scale-out to meet a wide range of price points and superior TCO points.



In the final installment on this topic, I will review the economic sweet spots for HC, in terms of workload, size, quantity and capability.