Infrastructure Economics #

Introduction #

Digital technologies are increasingly the raw material for how organizations drive operational efficiencies, improve products and services, and engage customers.

For enterprises that manage their own data center infrastructure, ensuring future services are available at the right time, in the right amount, at competitive costs depends upon the effectiveness of their inventory policies to optimally monetize hardware efficiencies and performance (HEP) while navigating inherent risks associated with highly perishable inventory that rapidly decays in economic value.

Perishable Inventory #

Computing resources cannot be saved, like stock inventory, to be used at a later time. A variety of business models have come and gone and even recycled — e.g., mainframe time sharing of the 1960’s and 70’s vs. public cloud providers — in a never-ending pursuit for how best to harmonize demand with resource availability.

Advances in hardware and OS virtualization, along with orchestration, have essentially decoupled hardware from software and rendered hardware fungible and workloads portable. This allows resources to be advantageously organized into logical collections composed of many physical server units, whose membership in any particular collection can be modified in real-time, in small increments, that better aligns resource availability to demand.

But this only makes higher resource utilization rates feasible. It doesn't guarantee them. Sustaining higher resource utilization depends on how much is made available for use. Determining what is needed, when, and “how much” is governed by an inventory policy that is, ideally, designed to orchestrate hardware, software, space, and power resource acquisition and deployment strategies that maximizes utilization of all resources and optimally minimizes total cost.

For enterprises, that is especially important since the collective cost for software, space, and power that are correlated to hardware consumption typically exceeds hardware cost by 2-10 times. Simply raising hardware utilization does little to ensure the lowest total cost.

Inventory Decay #

Because infrastructure resources are acquired to be operated, the variable of "how much" actually corresponds to two additional decision variables: The quantity of hardware that needs to be added and the quantity that needs to be removed from a collection (i.e., a resource pool). These exchanges are inherently asymmetric, and intended to inject higher levels of HEP into a pool by replacing relatively lower HEP units, thereby ameliorating the opportunity cost of economic decay.

A hardware unit's optimal lifespan is the time when peak economic value is realized with respect to the pool to which it is assigned.

Hardware Lifespan #

Most enterprises operate many types of resource pools in which the cost of software, power, and space resources - which are correlated to hardware consumption - collectively exceed the cost of hardware itself. Pool cost ratios (i.e., Kratio) typically range from 2-10 and are indicative of the amount of savings available using more precise estimation techniques to predict optimal hardware lifespan with respect to the unique cost structure of individual pools.

Attaining more precise lifespan forecasts requires solving an unknown function that provides a solution to the two embedded decision variables used to optimally determine "how much" inventory is necessary at a future time.   (This is some of the complexity that Zypr abstracts away.)

To illustrate how lifespan is impacted by different Kratio values, the break-even time of hardware lifespan can be approximated using the following formula:

Applying the formula to different cost structures illustrates the relative differences in the length of time required to recoup the cost of better hardware with respect to the expected displacement of other resource costs.

To appreciate how different cost structures drive different inventory policies, consider recent announcements by Microsoft to extend their equipment lifespan from four to six years, Google's move from three to four years and AWS's from four to five years for servers.

Major cloud providers have low Kratios because the cost of their software stack, while very high, is not correlated to hardware consumption. To illustrate its effect on inventory policy, assume a public cloud has a Kratio of 0.8 and and forecasts hardware performance will improve a very modest 15% annually. The breakeven lifespan is 5.8 years.

Although extending hardware lifespan does reduce hardware costs, which likely increases profit margin, extending hardware lifespan may also have the benefit of maintaining customers' service consumption rate — effectively a stealth price increase.

The table and chart below are based upon the above formula and provide context when forecasting future resource requirements and the implicit assumptions that are typically embedded in inventory policies and subsequent procurement decisions. For example, consider an enterprise that adopted a six-year server refresh policy similar to Microsoft. For that to approximate an optimal lifespan, the Kratio would need to be around three or less and server performance would only improve 5%, or less, annually for the next six years. That would be a breathtakingly low assumption and a third of what Microsoft's lifespan suggests.