possible reasons for increasing time of batch

Hi all!

I send jobs to a remote cluster machine and do a tic-toc around my batch command. Unfortunately the measured time increases a lot during a loop over 300 of similar jobs... Staring from 20s in the beginning I come to 200s already for job number 75...

Any ideas what could be the reason for that???

Thanks very much!

Rafael

[Rafael]

By the way, I tried to restart my local matlab and go on with the loop:

I again find times around 20s, so something happens on my local machine which slows me down...

No, it does not have the same effect if I "clear all". Then increasing starts at same already increased amount of time...

Answer 1

[Image Analyst]

It might depend on what you're doing. Are you doing anything with axes? I once noticed that when I was in an intensive loop repeatedly displaying new images in the same axes, it would get slower and slower until it was crawling along. Then I put in a "cla reset" just before I showed an image and after that it was uniformly speedy. Could that help you?

Answer 2

[Rafael]

Hi all. Thanks for the suggestion with these axes, but thats not relevant here I think.

I "simply" submit a function with a few basic parameters (no big data, only paths where to load the data from on the cluster) and a new CurrentDirectory via batch. The function then loads, processes and stores data on the clusters harddrives, so I dont keep the connection to the jobs.

Only thing what really bothers me is the incrasing amount of time my loop needs per job-submission.

(Cluster uses torque as jobmanager)

As a start, I would like to have some trick to uncover the bottleneck...