We've just released new versions of Analytic Solver® V2022 -- our product line for Excel -- and RASON® V2022 -- our cloud decision intelligence platform. Our theme for these releases has been "better performance and better resilience".
By "better performance" we mean faster solutions, especially for large optimization problems, but also faster interactive response time, and even faster startup of Microsoft Excel. By "better resilience" we mean that Analytic Solver does better with hi-res as well as limited-depth video monitors, and RASON does better with both extreme loads and "extremely odd" models.
Better Performance
In desktop Excel for Windows, we believe users will notice better performance right away. We've focused on minimizing the work that Analytic Solver must do when Excel starts up, and also minimizing the impact of work Analytic Solver must do to keep up with changes in your worksheet model, when you are not actively using the add-in's own Ribbon tab or Task Pane.
When solving an optimization model, in both Analytic Solver and RASON, we've focused on the time spent in "Diagnosing model" and "Setting up problem" -- this is where our PSI Interpreter algebraically analyzes your Excel formulas, determines how your objective and constraints depend on your decision variables, and generates the much simpler internal form that optimization solvers understand (a "Jacobian matrix" in general, an "LP coefficient matrix" for linear programming problems). Generally, the larger the model, the greater the performance improvement. But the improvement is even greater on models with long "chains" of formulas that depend on other formulas, sometimes to 10,000 or even 100,000 levels deep.
And of course, we've included latest versions of our large-scale Solver Engines, featuring the Gurobi Solver 9.5 -- the highest performance version ever of this popular Solver for linear and quadratic mixed-integer problems.
Better Resilience
In Analytic Solver, as mentioned above, we've focused on better rendering of dialogs on the wide range of video layouts and resolutions in use today. In both Analytic Solver and RASON, we've focused on better handling of models that "do crazy things", from trying to run millions of parameterized optimizations or simulations in a student exercise to using undocumented / unintentional Excel formula behaviors.
On the RASON Server -- where our ongoing "load" of models to be solved has grown considerably, from our spectrum of users (RASON itself, Analytic Solver Cloud, even basic Solver for Excel for the Web and for Google Sheets) -- we've focused on better end-to-end performance, even when solving many models from many users simultaneously, and better utilization of "cloud service main memory", enabling RASON to solve bigger models than ever, at the same time it solves large numbers of small models -- while ensuring that an "errant model" never impacts the performance of well-behaved models.
Some Other Enhancements
These enhancements were in response to widely-shared desires from our customers. But we've also included some small enhancements in response to individual customer requests. A feature called Freeze and Thaw, removed from Ribbon menu dropdowns in an earlier release since we saw little usage of it, has been restored on the Task Pane Tools tab. This is used to share Excel workbook models with other users who don't have Analytic Solver installed: "Freeze" saves all formulas containing PSI function calls in cell comments, "Thaw" restores them from cell comments to actual formulas.
Another small enhancement deserves a little explanation: As most customers know, Analytic Solver comes with five built-in Solver Engines, and also supports eight optional (extra cost) large-scale Solver Engines. All eight optional Solver Engines are included when Analytic Solver is installed, but they are "unlicensed" so they cannot be used for free. But you can get an idea of how a large-scale Solver Engine will perform on your model, before you actually buy a license, by performing a Test Run. This will show you whether -- and how fast -- your model is solved by the Solver Engine, but it won't deliver actual values for the optional decision variables or constraints. To make this more obvious, in V2022 in the Task Pane Engines tab, "(Test Run)" appears next to the names of Solver Engines that are not yet licensed, but are available for Test Runs.