Warning on ctrl-c: Pressing ctrl-c in MATLAB when connected to MATLAB*P will kill the MATLAB*P process, and will cause a lot of error messages to spew out. If this happens, exit MATLAB, and then rerun /share/startmp

MATLAB*P homepage

MATLAB*P Quick Start Guide

• Run
  

      /share/getresv X

  where X is the number of nodes you want (remember, each node has 2 processors). This script obtains the nodes through PBS.
  
  
  
  
• After getresv runs to completion, your will notice you are logged into a node. You can run MATLAB*P/MATLAB with or without X support. Use

      /share/startmp 

  to start MATLAB*P (and MATLAB) without X support,
  
  
  
  or, to start with X support (so you can do plots, use the MATLAB Desktop ...etc) first run
  

      xhost +beowulf.lcs.mit.edu 

  on your local machine to allow X clients to connect from beowulf.lcs.mit.edu. (Note that you do not do xhost +compute-x-y because the compute nodes live on a private (10.*.*.*) network and forward their network connection through beowulf.lcs.mit.edu.) For example, if you are working on Athena:
  

     athena% xhost +beowulf.lcs.mit.edu

  
  Then (back on the compute node), run

      /share/startmp your-hostname.mit.edu:0.0 

  (replacing your-hostname.mit.edu with the name or IP address of the computer you are sitting at).
  
  
  

Basic commands

• np - returns number of processes in the server
• A = randn(512*p) creates a real, 2d block cyclic distributed matrix
• A = randn(512*p,512) creates a real, row distributed matrix
• e = eig(A) takes the eigenvalue of A
• A(1,1) - examines the (1,1) element of A
• ... so on and so forth
  
  
• B = mm('sin',A); execute the "MultiMATLAB" mode of MATLAB*P
  
  On a non-cyclic distributed matrix A, mm starts up a MATLAB session on each of the node, and evaluate the string argument inside MATLAB, using the local pieces of the distributed matrix/matrices as arguments. See "help mm" within MATLAB for more information and examples.
  
  
  
  
• A = matlab2pp(a) - brings a local matrix a to the server and store the returned handle in A
• b = pp2matlab(B) - brings a distributed matrix B to MATLAB and store it in local matrix b

Example scripts: