Providing storage and compute resources in a one-size-fits all unit of scale has several advantages, but it can also present several challenges for customers. The term “no free lunch” is often used, and there really is no escaping it when it comes to the implementation of full stack technologies. Trade-offs are always made when a product is designed and brought to market. When it comes to the choices and trade-offs that the first generation of HCI product made, the choice was to offer a lower point of entry for both cost and form factor.
Sharing cores and combining resources allows for a low price point and low point of entry. In many instances we see first generation HCI products offered in the 35-45k price range (sometimes even lower). These solutions are geared towards the smaller ROBO space, and when you look at the requirements for many edge based solutions, it makes perfect sense to combine resources if the overriding factor is cost. Many times the tradeoff here is performance and the ability to run mixed workloads. Sometimes it can be a lack of scalablity or mixing of node types that will require a customer to create a secondary cluster if the site needs to scale, or forces the customer into only a single node type that may not be what they need in the future.
Many first generation HCI solutions leverage a CVM (Controller Virtual Machine) to provide storage resources and a management plane across the common form factor of a 2U 4 Node unit of consumption with a direct attached storage model. The CVM takes resources or ‘overhead’ from those individual nodes. This overhead can vary based on a number of factors. In some instances first generation HCI products can consume as much as 20-30% of overhead from the memory and compute that exists on each of those individual nodes. Once can see that as a solution scales, that amount of overhead can contribute to additional cost, not just to replace the resources that could be dedicated in a separated architecture, but in the costs associated with licensing that is attached to a CPU socket or cores. This is the “HCI Tax” that comes with a simplified, all-in-one approach to virtualized infrastructure. And that’s a story for another blog post.
…. To Be Continued