Welcome to Version 10 of Frontline Systems' full product line. V10 powerfully expands the ways you can develop and deploy applications of optimization and simulation. Here are several highlights:
- All Excel products support Microsoft Excel 2010 -- 32-bit and 64-bit!
- Full set of 32-bit and 64-bit Solver Engines for use inside and outside Excel!
- Ability to create models in Excel and solve them outside Excel!
Risk Solver Platform V10.x and its subset products Premium Solver Platform, Premium Solver, Risk Solver Premium and Risk Solver work with 32-bit and 64-bit Excel 2010, as well as Excel 2007 and Excel 2003. With V10.5, you can create an optimization or simulation model in Excel, build an application program with Solver Platform SDK that runs anywhere -- on an Intranet or Web server, even on Linux! -- loads your Excel workbook, and solves your optimization or simulation model at lightning speed!
Risk Solver Platform V10.5 uses a deeply parallelized software design to exploit the multi-core processors in modern PCs to analyze and solve larger models, faster than ever before.
Multi-Core Monte Carlo Simulation: V10.5 will exploit multiple processor cores to further speed its already-fast “vectorized” Monte Carlo simulation.
Multi-Core Simulation Optimization: V10.5 will also exploit multiple processor cores to speed up simulation optimization – usually the most compute-intensive method used to solve optimization problems.
Multi-Core Global Optimization: When you use the Multistart methods for global optimization with the GRG Nonlinear Solver or with the LSGRG, LSSQP or KNITRO Solvers, V10.5 will exploit multiple processor cores to run multiple nonlinear optimizations in parallel.
Multi-Core Nonlinear Optimization: Even when you’re solving a single smooth nonlinear optimization with the GRG Nonlinear Solver or with the LSGRG, LSSQP or KNITRO Solvers, V10.5 is significantly faster – it will exploit multiple processor cores to speed up gradient computations via parallelized automatic differentiation.
Multi-Core Mixed Integer Nonlinear Solutions: V10.5’s standard Branch & Bound algorithm, which is used by the GRG Nonlinear, LSGRG and LSSQP Solvers, has been parallelized to exploit multiple processor cores on mixed-integer nonlinear (MINLP) problems.
Multi-Core Non-Smooth Optimization: V10.5 provides parallelized function evaluation and parallelized local search searches to Solver Engines – and these services are exploited in a major way by our redesigned Evolutionary Solver (see below).
Solving Problems on Supercomputing Clusters or in the Cloud
Risk Solver Platform V10.5 and Solver Platform SDK V10.5 include built-in support for high performance computing (HPC) clusters running Microsoft’s new Windows HPC Server 2008 R2. Risk Solver Platform includes easy-to-use menus and dialogs for connecting to a compute cluster, using it to solve problems, returning solutions to the desktop, and working with the solutions in Excel – just as if all the computing had been done on the desktop. Both Risk Solver Platform and Solver Platform SDK can run on a cluster’s compute nodes, and solve optimization and simulation problems submitted by users.
With Windows HPC Server 2008 R2 Service Pack 1, users who don’t have a large compute cluster in their company can create one “on the fly” by spinning up compute nodes as virtual machines on Windows Azure, and running Solver Platform SDK on those nodes to solve optimization and simulation problems. Once the problems are solved, the virtual machines are “spun down,” the user pays only for the time he or she uses them, and many IT maintenance issues simply disappear.
Several of the built-in Solvers V10.x feature algorithmic improvements that speed solutions. Many models will solve faster in V10.x.
LP/Quadratic Solver includes a wide range of improvements in presolve, preprocessing, cut generation and heuristics for LP/MIP problems. These methods are now automatically selected under overall user control – making them much easier to use. Performance is very model-dependent – but overall, you can expect a significant speed improvement on your LP/MIP problems. We expect many user models will solve twice as fast as in V9.0.
GRG Nonlinear Solver uses parallelized automatic differentiation, parallelized Branch & Bound for MINLP problems, and parallelized Multistart start for global optimization problems to greatly improve performance. On a quad-core PC, we expect many user models will solve twice as fast as in V9.0.
Evolutionary Solver has been redesigned to deeply exploit the search possibilities afforded by multiple processor cores. In addition to improvements in its algorithms, ‘balance of effort’ heuristics, and stopping rules -- which are available in all subset products for optimization -- the Evolutionary Solver in Risk Solver Platform and Premium Solver Platform now uses a parallelized (and ‘vectorized’) global search and a parallelized local search to find much better solutions in the same time. Performance is very model-dependent – but on a quad-core PC, we expect solutions at least twice as fast on many user models, and some models will see better solutions at speeds 10 times faster or more.
Plug-in Solver Engines Offer Best-Ever Performance with 10.5 upgrade
V10.5 also includes major new releases of several large-scale Solver Engines that “plug into” Risk Solver Platform, Premium Solver Platform and Solver Platform SDK. The Gurobi Solver Engine V10.5 now solves quadratic (QP) and quadratic mixed-integer (QP/MIP) problems with remarkable speed. The XPRESS Solver Engine V10.5 offers a typical 50% performance gain over its previous version on linear, quadratic and mixed-integer models. The KNITRO Solver Engine V10.5 offers industry-leading performance on large-scale non-linear models, and takes advantage of the Intel Math Kernel Library (MKL) to exploit multiple processor cores and advanced features of Intel’s latest processors.
Managing Solutions for Multiple Optimization and Simulation Problems
Users who build optimization or simulation models frequently want to solve many instances of these models with different data or parameters. For some time, Frontline’s Solvers have made it easy to solve multiple problem instances, but as with other software, the final solutions were obtained in memory and had to be used immediately for analysis or creating reports. V10.5 introduces an XML-based solution file that makes it easy to solve multiple problem instances on any machine, save, transfer and load solutions, and use solutions for analysis and reporting on any machine where Frontline’s software is running.
Use RSP's powerful features for sensitivity analysis of your Excel model, before you even start an optimization or simulation model. Identify model parameters with the most impact on your computed results – simply select any formula cell, and choose Parameters – Identify from the Ribbon to quickly find the input cells with the greatest impact on this formula, ranked and shown in a Tornado chart. You can choose some of these input cells to serve as Sensitivity Parameters, and then produce reports and charts that show the impact on computed results of varying these parameters over a range you specify. You can also turn these parameters into Simulation Parameters or Optimization Parameters, and produce reports and charts of multiple simulations and optimizations.
Create decision trees on your Excel spreadsheet. Using the Ribbon, you can easily create decision nodes and branches, event nodes and branches, and terminal nodes. The tree is drawn in graphic form on the spreadsheet; standard Excel worksheet formulas compute ‘rollback’ values at each node, and the best-choice value at the root node, based on either expected value or utility function (certainty equivalent) criteria. With a Ribbon choice, you can graphically highlight the optimal path through the tree. Since all computations for decision trees are performed via standard Excel worksheet formulas, you can use decision trees in your simulation and optimization models.
New in Risk Solver Platform V10.x is a facility for multiple discriminant analysis (MDA). Discriminant analysis is a statistical technique for analyzing a data set, consisting of observations of one or more variables, and classifying the observations into different groups. It has many applications in financial services and consumer marketing, among other fields. Risk Solver Platform implements both linear and quadratic methods for MDA.
Improved Support for Shared Probability Distributions
In V10.5, Frontline continues its leading-edge support for shared probability distributions, used in simulation and stochastic optimization applications, stored in the DIST 1.1 format – an emerging standard now supported by other software vendors such as Oracle and SAS. In Risk Solver Platform V10.5, it is easier than ever to produce and consume arrays of DIST-based probability distributions in a model, and use them in ultra-fast Monte Carlo simulations in Excel – 10 to 100 times faster than most other software.
Risk Solver Platform V10.x and all of its subset products incorporate a new license manager, from Reprise Software, that replaces the Sentinel license manager that Frontline used in V9.0 and earlier versions. For customers with Flexible Use (“concurrent user”) licenses, we have a new License Server to work with this license manager. The Reprise license manager offers simpler and more reliable trial licenses, plus the considerable convenience of Internet-based license activation. With V10.5, we’ll no longer have to ask you for a lock code and email you a license code that you have to copy and paste; we can simply give you an “activation code” – about the length of a credit card number – and you can use this to activate your own license through our Internet server.
In all V10.x products, you can get quick online Help for every Solver Result message and error message in the Output pane, every Platform option and Solver Engine option, and every element of your simulation model, optimization model, or sensitivity analysis or decision tree element in the Task Pane Model tab. V10.5 also has a considerably more useful startup screen that appears only when you visit the Risk Solver Platform tab on the Ribbon, and then only when you’re using a trial license.
Using the Help - About menu choice, you can switch between Risk Solver Platform, Risk Solver Premium, Premium Solver Platform, Premium Solver, and Risk Solver. You can run any product that equals, or is a subset of, the product for which you have a license. This is especially useful when evaluating our software, since you can 'experience' exactly what features and performance are available in each subset product.
Solver Platform SDK V10.0 now includes the full power of PSI (Polymorphic Spreadsheet Interpreter) that was previously available only in Risk Solver Platform and Premium Solver Platform. The SDK can read your Excel workbook, run single or multiple Monte Carlo simulations, and use any of Frontline's twelve built-in and plug-in Solver Engines for optimization of large-scale models! It's easy to obtain new data -- from a database, another application, or an end user -- and use it to update your Excel model, and re-solve. You can even solve an Excel workbook model on Linux!
With the PSI Interpreter, Solver Platform SDK gains multiple new capabilities, such as the ability to analyze and transform your Excel model, high-speed vectorized Monte Carlo simulation and simulation optimization, use of the full spectrum of Excel formulas and the full Excel function library (even the new Excel 2010 functions), the power to solve large-scale linear models with uncertainty using stochastic programming and robust optimization methods, and more.
You can still create your optimization or simulation model in a programming language, from C/C++ to C# and VB.NET, Java and MATLAB, in Solver Platform SDK V10.5, and you can still load models defined in MPS format, LP format, or OSiL format. This makes the SDK an amazingly powerful and versatile tool for development and deployment of your models. It's available for 32-bit and 64-bit Windows and Linux, and it accepts any of Frontline's eight plug-in Solver Engines, 32-bit and 64-bit versions.