Print

[keymaker]

Hi:
I have a problem with JPPF 3.0 and intel core i7 processors. I´m running an application in a grid of core i7 computers. I can´t achieve that the cores of the processors work in a parallel balanced way. Also the cores never reach 100% of use.  The load balancing algorithm that i use is manual and proportional. I modified the amount of tasks that both algorithms must distribute to the processors and is the same thing.
i saw a previous post in this forum and i did this but it doesn´t work

# specify the algorithm
task.bundle.strategy = proportional
# name of the performance profile to use
task.bundle.autotuned.strategy = optimized_strategy
# define the algorithm parameters
strategy.optimized_strategy.performanceCacheSize = 3000
strategy.optimized_strategy.proportionalityFactor = 2

Thanks in advance

[lolo]

Hello,

From the property names you provided, it looks like you are using the property names for an older version of JPPF.

Here's what I would configure in the server configuration:

Code: [Select]

jppf.load.balancing.algorithm = proportional
jppf.load.balancing.strategy = test

# "proportional" profile
strategy.optimized_strategy.performanceCacheSize = 3000
strategy.optimized_strategy.proportionalityFactor = 1
strategy.optimized_strategy.initialSize = 1
# initial mean time expressed in nanoseconds
strategy.optimized_strategy.initialMeanTime = 1e9
A "proportionality factor" greater 1 generally does not provide good results, so I recommend setting it to 1.

Additonally, the average CPU usage will highly depend on the computational weight of your tasks. For instance, if you have very short lived task, which only take a few milliseconds to execute, then it's possible that a large part of the total execution time will be spent in I/O between the node and server, rather than pure CPU-bound processing.

I hope this clarifies.

Sincerely,
-Laurent

[keymaker]

hi i did all that and still doesn't work the cpu doesn't use more that the 50%. I run various clents simultaneously and the behaviour doesn't change. The task lifecycle son very big so must be problems with the load balancing algorithm and the i7 processors. I hope any help from you. Thanks

[lolo]

Hello,

In my previous post, I made a mistake for the load balancer configuration. Instead of "jppf.load.balancing.strategy = test" you should read "jppf.load.balancing.strategy = optimized_strategy", sorry about that.

Also, as I mentioned, the actual CPU usage will depend on what's in your tasks. Could you tell us how many tasks you typically have in a job, how many nodes you have, and how long each task takes to execute?

Thanks,
-Laurent

[keymaker]

Hello

Regarding the number of cores that use are two i7 nodes and the server  is also an i7 computer. I lean on the example of matrix multiplication that brings JPPF, change a few things to fit my problem. The matrix has an order of 5000x5000, each task has 100 rows of matrix "A". Finally, I have 50 tasks to execute on the i7.

Today I will try again with a configuration that you suggested me and soon I inform you if it worked