{"id":5458,"date":"2014-08-01T09:00:23","date_gmt":"2014-08-01T13:00:23","guid":{"rendered":"https:\/\/blogs.mathworks.com\/pick\/?p=5458"},"modified":"2014-07-31T20:02:13","modified_gmt":"2014-08-01T00:02:13","slug":"time-parallel-for-parfor-loops","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/pick\/2014\/08\/01\/time-parallel-for-parfor-loops\/","title":{"rendered":"Time Parallel FOR (PARFOR) Loops"},"content":{"rendered":"
\r\n \r\n

Sean<\/a>'s pick this week is Par Tic Toc<\/a> by Sarah Wait Zaranek<\/a>.\r\n <\/p>\r\n <\/introduction>\r\n\r\n

I run a fair number of computationally expensive operations and often use parallel for (parfor)<\/a> loops to speed them up. This utility interested me mainly because it shows how work is distributed across the workers and\r\n provides worker utilization rates.\r\n <\/p>\r\n

First, I will make sure a parallel pool is open using gcp()<\/tt><\/a> or \"Get Current Pool\". This opens the headless MATLAB processes that a parfor loop distributes iterations to.\r\n <\/p>

pool = gcp<\/pre>
\r\npool = \r\n\r\n Pool with properties: \r\n\r\n            Connected: true\r\n           NumWorkers: 2\r\n              Cluster: local\r\n        AttachedFiles: {}\r\n          IdleTimeout: 90 minute(s) (89 minutes remaining)\r\n          SpmdEnabled: true\r\n\r\n<\/pre>
We'll do a simple example with a loop that has a random pause.  The pause is a proxy for an operation that takes a varying\r\n      and unpredicable amount of time.\r\n   <\/p>
n = 50;\r\np = Par(n); % Build the Par timer<\/span>\r\nparfor<\/span> ii = 1:n\r\n    Par.tic; % start timer<\/span>\r\n    pause(randi(5)) % Pause for up to five seconds<\/span>\r\n    p(ii) = Par.toc; % measure elapsed time<\/span>\r\nend<\/span>\r\nstop(p); % all done<\/span>\r\nplot(p);<\/pre> 
The parfor loop has done a good job distributing the calculations between the two workers.  The more highly utilized worker\r\n      only has one additional iteration that the other.\r\n   <\/p>\r\n   
There are a few other neat tricks in here as well that I found from looking through the code.<\/p>\r\n   
\r\n      
    \r\n         
  1. If you right click on the axes with each iteration's time, it gives you context menu to select the scale.<\/li>\r\n         
  2. Since the Par class inherits from handle, you get those visible methods.  Sarah hides those from methods()<\/tt><\/a> by making their signature hidden.\r\n         <\/li>\r\n         
  3. Sarah has also provided a full set of documentation on the capabilities of this utiltiy as well as examples.<\/li>\r\n      <\/ol>\r\n   <\/div>\r\n   
    Comments<\/a><\/h3>\r\n   
    Give it a try and let us know what you think here<\/a> or leave a comment<\/a> for Sarah.\r\n   <\/p>