Support:Documents:Manual:Distributed Computing with COMKAT

From COMKAT wiki
Jump to navigation Jump to search

What is Matlab Distributed Computing Server (MDCS)

The purpose of using Matlab Distributed Computing Server (MDCS) is to reduce computational time for data-intensive problems. It is executed by MATLAB and Simulink based apllications on a computer cluster, and it is available for all hardware platforms and operating systems. More detail introduction about MDCS can be found in this link.

How can MDCS help COMKAT

In kinetic modelling, there are several time-consuming computations. For example, it takes several days to perform a pixel-wise estimation for kinetic parameters. One solution to reduce the computational time is to use MDCS. This example can be found in the following link. One of the other applications is about setting initial conditions for estimating kinetic parameters. Generally, the accuracy of estimated kinetic parameters is easily afftected by their initial conditions. Therefore, it is important to choose an appropriate initial guess. However, there is no standard rule to find an appropriate initial guess. One alternative is to perform parameter estimation with different initial conditions and to calculate the mean kinetic rate constants from these different conditions. However, this method is limited by its computational time. Fortunately, its computational load can be reduce by using MDCS. So, MDCS can help COMKAT to reudce computational time for data-intensive problems.

Setting Matlab Distributed Computing Server (MDCS)

To start parallel computing, user must install MDCS that is described in this document link. To get appropriate instructions, user should click the document link above and fill out the information about your cluster computer's platforms, Matlab end user's platforms, installation location, scheduler and licensing, and there should be four stages for installing MDCS. The below paragraph is a brief summary for installing MDCS. The customized document generated by clicking on the document link above gives details on how to set up the user's computer. Please review that document for details and screen snapshots.

Following the below three stages if it is the first time you are building MATLAB Distributed Computing Server

Stage 1: Install MATLAB Distributed Computing Server

Stage 1 is to install MDCS on the head node and worker nodes and then start the license manager on the head node. After installing MDCS, user should start the license manager on the head node. On the head node, navigate to the matlabroot\flexlm folder and start the Macrovision LMTOOLS utility by double-clicking the lmtools.exe file. Then, click the Start/Stop/Reread tab and click the Start Server button. Look for the status message “Server Start Successful” in the bottom of the LMTOOLS window.

Stage 2: Configure Your Cluster for use with MathWorks Job Manager

Stage 2 is to renew mdce services on all nodes. After updating mdce services, user can start the job manager and workers. To open Admin Center, navigate to the folder: matlabroot\toolbox\distcomp\bin and then execute the file:admincenter.bat. Then, follow the document in the stage 2 to add new job manager if it is the first time you are using MDCS. Note: To use Admin Center, users must run it on a computer that has direct network connectivity to all the nodes of your cluster.

Stage 3: Install Parallel Computing Toolbox

Stage 3 is to install parallel computing toolbox if you will write Matlab applications.

Setting Up Client (User's Computer) for Matlab Distributed Computing

Following the below stage if it is the first time you are using MATLAB Distributed Computing

Stage 4: Test Your Parallel Computing Environment

The first step in this stage is to define a user configuration. In brief, the steps for a Windows client are to click on the Parallel menu item (to the right of File, Edit, ... in the MATLAB window). If this is the first time you are using MATLAB Distributed Computing, click Manage Configurations.... In the Configurations Manager window that opens, click File On the File -> New menu item and select the desired jobmanager type. To use MATLAB's built-in manager, select jobmanager. Define a name and other properties for the configuration including the Job manager host name (e.g. an IP address) and the job name manager. On the jobs tab, specify the maximum and minimum numbers of workers. Click OK/save. The second step is to verify the network connection and validate the configuration. On MATLAB 2008b or newer you may validate the configuration: Click on the Parallel -> Manage Configurations menu item to start the Configurations Manager and select your configuration in the dialog box listing. Click Start Validation. If your validation does not pass, contact the MathWorks install support team listed in the document.