Author |
Message
|
zpat |
Posted: Mon Mar 15, 2010 8:57 am Post subject: WMB and AIX WorkLoad Manager (WLM) |
|
|
 Jedi Council
Joined: 19 May 2001 Posts: 5866 Location: UK
|
Has anyone used WLM on AIX to control different WMB workloads?
I would be interested in any suggestions on how to approach this, so that we could limit certain message flows to using a certain max proportion of system resources (CPU etc). |
|
Back to top |
|
 |
zpat |
Posted: Tue Mar 16, 2010 1:58 am Post subject: |
|
|
 Jedi Council
Joined: 19 May 2001 Posts: 5866 Location: UK
|
WLM needs to identify different processes to assign them to workload classes.
Has anyone been able to identify different execution groups in this way?
This could provide a neat way to both control and charge back WMB resource usage.
The process name looks like this in ps -ef
Quote: |
DataFlowEngine XXXXXXX 419c9f8b-2101-0000-0080-d3d14b9bbfc6 AAAAAA |
Where AAAAAA is the EG name. |
|
Back to top |
|
 |
zpat |
Posted: Tue Apr 13, 2010 5:30 am Post subject: |
|
|
 Jedi Council
Joined: 19 May 2001 Posts: 5866 Location: UK
|
WLM can classify a process based on various rules, ie. userid, groupid, application path name, wlm tag.
I want to be able to put different WMB execution groups (from the same broker) into different WLM classes for performance isolation.
I can't because EG processes in the same broker have the same userid, group, and application path (the parameters to DataFlowEngine cannot be used as a WLM rule).
What I need is either for WMB to set a WLM tag value (equal to the execution group name or message flow name) or to be able to run different EGs under different AIX Userids.
Anyone know if either of these can be achieved? It would be great if IBM could make two of their products work together (WLM and WMB).
In any case it would be helpful for security isolation to be able to run different EGs within one broker under different Userids. |
|
Back to top |
|
 |
zpat |
Posted: Thu Apr 15, 2010 11:11 pm Post subject: |
|
|
 Jedi Council
Joined: 19 May 2001 Posts: 5866 Location: UK
|
I am clearly talking to myself here.
However the solution is to use the wlmassign command to manual assign the EG processes to a WLM class.
A small script can be run to automate this. Not ideal but good enough.
I am surprised no-one else seems to be interested in WLM with WMB. |
|
Back to top |
|
 |
fjb_saper |
Posted: Thu Apr 15, 2010 11:26 pm Post subject: |
|
|
 Grand High Poobah
Joined: 18 Nov 2003 Posts: 20756 Location: LI,NY
|
zpat wrote: |
I am clearly talking to myself here.
However the solution is to use the wlmassign command to manual assign the EG processes to a WLM class.
A small script can be run to automate this. Not ideal but good enough.
I am surprised no-one else seems to be interested in WLM with WMB. |
You seem to be forgetting a few things here....
What uniquely identifies the EG looking at ps -ef is:
Code: |
ps -ef | grep <BROKER> | grep <egname> |
The pid can change. For example, just identify it and kill it with -9. The broker will reinstantiate it with a different pid.
Now all you need is to run multiple brokers on the same machine with the same eg names (like DEV + QA)...
I understand your question about the workload. If it is just about charge back why not make that be part of the MQ world, logging each message to a DB with a charge back on the accounting field? This of course implies you'd know how to set it when the input node is HTTP or SOAP....
Have fun  _________________ MQ & Broker admin |
|
Back to top |
|
 |
zpat |
Posted: Fri Apr 16, 2010 3:11 am Post subject: |
|
|
 Jedi Council
Joined: 19 May 2001 Posts: 5866 Location: UK
|
Of course the pid will change, but the manual assignment script can be run regularly with a cron job to (re)set the correct WLM class.
The selection of the process can be done by matching both broker and EG name.
Chargeback may become a requirement - right now I am trying to ensure all the different applications get their fair share of the resources. |
|
Back to top |
|
 |
|