More Precise Forecasts #

Background #

Modern technologies and architectures have made workloads highly portable, hardware nearly fungible, and resources composable. On-premise hardware resources can now be easily organized and reorganized in countless ways, over their service life, to satisfy service level needs as well as minimize the total cost of infrastructure services.

Zypr employs a far more advanced mathematical model unlike any TCO model or calculator. This gives planners the predictive precision to quickly evaluate feasible alternatives and cost-effectively right-size services — regardless of the resource type. An optimal solution for one resource type is an optimal solution for all resource types because Zypr simultaneously solves for hardware, software, power and space resources by fusing a broad range of capabilities that delivers coherent, actionable forecasts.

Complex Decision Variables #

• Projected service demand growth rate
• Current resource inventory
• Different computing efficiencies among existing hardware inventory
• Expected improvement rates in hardware computing efficiency
• Software terms and counting rules
• How efficiently resources will be utilized

Applied Optimizations #

• Derivative-based optimization
• Optimal assignment

Time-Scalable Forecasts #

• Resource lead-times that range from weeks to years
• Resource commitments based on a "predicted hardware plan"
• Uncertainty of future hardware computing efficiencies

Event-Based Simulation #

Zypr is a discrete-event simulator with specialized engines for integrating derivative-based optimization and complex software usage terms and counting rules. Planners describe an infrastructure service using a Pool Resource Model. The simulator automatically organizes the event-base interval structure for evolving a service's resource composition over a preferred time horizon and returns a Resource Requirements Forecast.

Solution Construction #

Zypr enables two types of simulations: 1) Optimal, and 2) Fixed-Time. Both types use the same event-based model paradigm, but invoke different variables and rules to solve for two different inventory policies.

Optimal simulations do not require the model to specify a preferred fixed-time target for server refresh nor the chronological age of existing hardware inventory, whereas Fixed-Time simulations require both. How each generates a forecast therefore differs.

For both simulation types, Zypr will automatically organize both defined and auto-calculated (e.g., add hardware capacity at peak utilization boundary) discrete events into a sequence of transactions by event time, as shown in the illustration below. Zypr evaluates the feasible states when transitioning from one time-interval to the next based upon the proposed state change as derived from continuous-form decision variables and one or more discrete events.

For an optimal simulation, the feasible solution space at each interval represents unknown hardware add/removal rates. Zypr therefore will evaluate each solution space, f, in order to find the sequence of interval-based hardware add and removal combinations that minimizes total resource costs for the specified time horizon t. The Resource Requirements Forecast is then derived from this least-cost sequence.

In contrast, a Fixed-Time simulation specifies a predetermined server removal rate, which creates a known feasible solution space size of one at each interval. Since the removal rate is known, Zypr solves the unknown server adds rate. Because the solution space is limited to one, the solution is unlikely to be the lowest total cost solution when hardware cost comprises less than 50% of a pool's total resource cost.