Increasing the speed of Decision Making

One of the key issues facing simulation modellers is the time it takes to run simulation models. Typical models may take 40 minutes or more to run a single replication, and with a typical scenario needing to be run for at least 10 replications, the results for any scenario can be a source of significant delay. When you then realise how many scenar- ios a user may like to run, not to mention the runs used to test and validate a model, the workload can be significant.

Saker Solutions have worked with Brunel University on an R&D project looking into Distributed Simulation. The first phase of this work has resulted in the development of a product to enable simulation users to easily distribute scenarios across a local Grid achieving a near linear improvement in response performance.

Figure 2. SakerGrid Performance.

SakerGrid is available either as a Desktop Grid or as a Virtual Grid. As a Desktop GridSakerGrid utilises the corporations network of simulation software and will distributeindividual replications, (or blocks of replications where the model run time is very shortand therefore it is more efficient to manage multiple replications sequentially) to suita-bly licensed machines on the network. SakerGrid will therefore work happily in a multi-software environment distributing models to machines which have the appropriate simulation software and where appropriate, libraries installed. Further options allow the Desktop grid to use either stand-alone PC's or a network cluster and indeed this can be time dependent allowing SakerGrid to either utilise desktop PC's at any time or only overnight.

The VirtualGrid allows users to submit jobs to the Saker Managed Grid which is ideal for users who have limited access to simulation software. With Flexsim models for example this allows users to submit a model and get access to 50 times response improvements from a secure environment.

There are 3 elements to SakerGrid

Client
The Client application allows the user to interact with the Grid, submitting jobs and collecting results. Through the client application the users can set the priority of their jobs and see the status of their jobs on the Grid.

Manager
The Manager functionality dispatches jobs across the network. The Manager util- ises scheduling logic to distribute jobs according to priorities and rules assigned by the Grid administrator. The administrator can not only see the status of jobs on the Grid but also has the capability of moving jobs up or down the Grid’s queued jobs or even deleting

jobs in the queue entirely. The Manger also gives a view of the performance of the disparate hardware on the Grid and allows the administrator to set options which may be either time or performance dependent.

Worker
The Worker application sits on each machine in the grid and communicates with the Manager information relating to its capability (licensed software), availability, results from any replications and any failure issues whether caused by a failing in the model or the infrastructure.