The Cephei kernel implements a dynamic graph dependency tree for calculation, to trigger re-calculation when a dependant cell is changed. Calculation of independent paths through the tree are executed in parallel to use all available CPU cores on the computer.

Within Cephei.XL the kernel maintains the dependencies between cells without relying on Excel to know which cells are dependent on each other. The Cephei.XL tree dependencies are used to generate F# source code in the correct order.

Compiled models re-create the same dependency tree as the Spreadsheet model when loaded providing the same dependency tracking behaviour.

In this sample dependency tree any change to “future 1” (a futures price) will trigger the re-calculation of the Bond NPV price, which can trigger P&L

Whilst a calculating through a dependency graph uses more CPU time than a traditional imperative approach, when independent calculations are performed in parallel the elapsed time is lower because more CPU cores are used. With standard Xeon processors holding 18 cores (36 cores for a cheap trader workstation) parallelism wins.