Print

[ygzx_114]

hello
     1.  I found the default algorithm is proportional whit strategy "test". it will calc 1 tasks first, and according to the calc time,  distribute tasks to different nodes. if I want each node calc 2 tasks first, then evalute the calc ability. which parameter can I adjust?
    2.  when I ran a job with 4 nodes connects to the server, I found that one node calc half of the tasks, and some nodes calc only a few tasks.  how can I adjust the profile to improve this situation. or where can I find the document abount adjust the Loading balance strategy profile.
    3.  When I change the  load balancing algorithm to manual and ant the drive,then restart the drive, but at the JPPF Monitoring and Administration tools, I can see that the algorithm is still proportional, How can I change the  load balancing algorithm ?

    Thanks for your time,
    DoubleT
   

[lolo]

Hello,

1.  ... if I want each node calc 2 tasks first, then evalute the calc ability. which parameter can I adjust?

This is not yet possible in JPPF 2.5. It will be in JPPf 3.0 which is due soon, see this feature request: 3315028 - Configurable initial bundle size for proportional algorithm

2.  ...  how can I adjust the profile to improve this situation. or where can I find the document abount adjust the Loading balance strategy profile.

You can take a look at this old section of the documentation: http://www.jppf.org/doc/v2/index.php?title=JPPF_Performance
It's a little out of date, but the description of the "manual", "proportional" and "autotuning" algorithms is still valid.
Don't forget that for "proportional", the number of tasks sent to each node is computed based on the past performance of the node. This implicitly includes the node's hardware capabilities (CPU/cores) as well as the quality of the network connection between node and server. For instance if you have 2 identical node, but the network connection is slow between one of them and the server, then the server will send more tasks to the node with the fastest connection.

3. ... but at the JPPF Monitoring and Administration tools, I can see that the algorithm is still proportional, How can I change the  load balancing algorithm ?

In the load balancing panel, there is a "Refresh settings" button, have you used it? It should refresh the values in the admin console with those in the server.

Sincerely,
-Laurent

[ygzx_114]

Hello Laurent，
   Thanks for your replying.
     the third question. I have used the "Refresh settings" button in the load balancing panel.
      I tried to restart the JPPF Monitoring and Administration tools, It's also no used.
     Additional, I changed the  load balancing algorithm  to "manual" in the load balancing panel, and used "Apply settings" button, then "Refresh settings", the load balancing algorithm back to "proportional".

    Sincerely,
 -DoubleT

[lolo]

Hi DoubleT,

Indeed, this is a bug in the server's management MBean, which always returns the load-balancing parameters read from the server's configuration, instead of those set from the admin console.
So the information displayed in the console in incorrect, however the new parameters you set from the console are really taken into account by the server.
I registered a bug for this:  3468321 - JMX does not return the latest load-balancer information.
I will let you know when it is fixed, and make a patch for JPPF 2.5.5.

Sincerely,
-Laurent

[lolo]

Hello again,

This bug has now been fixed, and a patch is available there.
Please let us know if this works for you.

Sincerely,
-Laurent

[ygzx_114]

Hello Laurent,
      Thanks for fix the bug so quickly.
      I used the patch and it works now.
     There still two questions, I changed the  parameters in the load balancing panel, and used "Apply settings" button, then "Refresh settings", I can see it's changed in the console. But It's not change in the driver config file, when I restart the driver and "Refresh settings", the parameters will go back. Does this normally？
     In the Topology panel of the console, the number of Task Executed is always 0, while it record task executed before.
       Sincerely,
 -DoubleT

[lolo]

Hi DoubleT

For you first question, yes it is normal. The configuration file is never changed. This si the way the code is designed, because there are cases when it's not possible to change it, for instance when the config file is within a jar file (or a war file for the Tomcat connector), or when you use a configuration plugin (with -Djppf.config.plugin=my.own.ConfigurationSource).

For the other issue, I cannot reproduce this behavior. Are you seeing any error or exception in the admin console log file or in the node log file?

Thanks,
-Laurent

[ygzx_114]

Hello Laurent,
      I don't see any error and exception in the admin console log file and in the node log file. I'm confused about this these days.

      There is another question about load-balancing algorithm, In my opinion, the algorithm which can makes all nodes work until the job end is best. So the "Manual" algorithm with the "size = 1" is the best algorithm. Dose this right?
      I had a try with the  "Manual" and "proportional" algorithm, "proportional" algorithm took 2720 seconds. more then "Manual"  algorithm, the Statistical data of "Manual"  algorithm as follows:
      strategy.test.size       1         2             5           10         15          20
             time               1610s   2049s     2102s     2377s      2350s      2439s
     
Sincerely,
 -DoubleT