| Author |
Message
|
| Bobbo |
 Posted: Fri Aug 19, 2005 6:53 am Post subject: Creating and Starting Large Numbers of Instances |
 |
|
Acolyte
Joined: 17 Jun 2002
Posts: 50
Location: Buffalo, New York
|
We are creating and starting 7000 new instances a day. The problem is that it is taking about 45 minutes to load 1000 instances and we only have about a 2 hour window. The process model contains a number of UPES activities that call Message Broker Flows, query a database, and return. Once the UPES activities have completed, work items are generated for manual activities. We have followed all the best practices, etc. from IBM for performance tuning, etc.
I was curious as to how others handle a situation like this. I have read the posting about creating an "Instance Server." However, that will not work for our situation because we need the instances started before users arrive to do the work.
I'm searching for creative ideas around loading this many new instances a day. Does anyone have any experience with this? If so, what is your platform? Is this considered a lot or should WF be able to easily handle this volume?
Thanks,
Bob |
|
| Back to top |
|
 |
| jmac |
| Posted: Fri Aug 19, 2005 10:38 am Post subject: |
|
|
Jedi Knight
Joined: 27 Jun 2001
Posts: 3081
Location: EmeriCon, LLC
|
Bob:
You must have other workflow processes that are already running right? Your numbers are pretty low. It is 3.9 BWU to create and start an instance, I would think you would see better response than you are reporting.
What else can you tell us about the process model and about the environment?
_________________
John McDonald
RETIRED |
|
| Back to top |
|
|
| Bobbo |
| Posted: Fri Aug 19, 2005 11:09 am Post subject: |
|
|
Acolyte
Joined: 17 Jun 2002
Posts: 50
Location: Buffalo, New York
|
Thanks for the reply.
All parts are running the z/OS.
Process model begins with 3 syncrhonous UPES activities, and then goes into a block that contains 1 additional synchronous UPES and 3 NOOPS before it hits the manual activities. The UPES activities are each launching a Message Flow that is doing a simple SELECT from a single DB2 table. Auditing is to MQ, using filters to minimize the number of records written. Clean Up Server and Scheduling Server are OFF during load.
We've tried running the load with 2, 3, and 4 Execution Servers. When running 3 and 4 Exe Servers, we are maxing out the CPU.
Our hunch is that the bottleneck is the Exe Server(s) reading from the EXE and XML queues. Any way to tune this? Any other possibilities?
Thanks,
Bob |
|
| Back to top |
|
|
| jmac |
| Posted: Fri Aug 19, 2005 11:27 am Post subject: |
|
|
Jedi Knight
Joined: 27 Jun 2001
Posts: 3081
Location: EmeriCon, LLC
|
Bob:
I have no experience running MQWF on Zos. Have you had a look at WP01? It seems to me that you are really not putting all that much of a load on the system. The way I read it the the lowest rate I see is 15,744 BWU/hour, there are others showing 500,000+, so a lot depends on what hardware you are running. You indicate 1000 instances is taking 45 minutes to load. To me that is about 4000 BWU for the process starts only, but it still seems like it should only take 15 minutes to start 1000 instances.
You might want to consider opening a PMR.
GOOD LUCK
_________________
John McDonald
RETIRED |
|
| Back to top |
|
|
| Bobbo |
| Posted: Fri Aug 19, 2005 11:37 am Post subject: |
|
|
Acolyte
Joined: 17 Jun 2002
Posts: 50
Location: Buffalo, New York
|
Yes, we have looked at WP01. IBM should release an updated version of that support pac.
We are working with IBM on this issue, but they too are having trouble finding a qualified WF 3.4 on z/OS performancing tuning expert.
By the way, the instances start rather quickly (in about 5 minutes, based on Start Time of Process Intances). It's the UPES / NOOPs that seem to be taking all the time.
Bob |
|
| Back to top |
|
|
| jmac |
| Posted: Fri Aug 19, 2005 12:05 pm Post subject: |
|
|
Jedi Knight
Joined: 27 Jun 2001
Posts: 3081
Location: EmeriCon, LLC
|
I am not getting something here... Are you saying that it is taking 45 minutes for these 1000 instances to complete? Or is it just the 1000 process starts that is taking the 45 minutes?
If you want to send me your FDL, I can take a look and see if I can spot anything for you... Most likely, it has already been looked at by IBM, but I can supply another pair of eyes if you want.
You can send the FDL to my email address rather than posting to the forum.
_________________
John McDonald
RETIRED |
|
| Back to top |
|
|
| Bobbo |
| Posted: Fri Aug 19, 2005 12:09 pm Post subject: |
|
|
Acolyte
Joined: 17 Jun 2002
Posts: 50
Location: Buffalo, New York
|
It is taking 45 minutes from the first work item being created in the first instance to the last work item being created in the final instance.
I'll see if I can send you the FDL (legal).
IBM has not yet seen the FDL. |
|
| Back to top |
|
|
| jmac |
| Posted: Fri Aug 19, 2005 12:19 pm Post subject: |
|
|
Jedi Knight
Joined: 27 Jun 2001
Posts: 3081
Location: EmeriCon, LLC
|
| Bobbo wrote: |
| It is taking 45 minutes from the first work item being created in the first instance to the last work item being created in the final instance. |
OK.... that makes a little more sense.
_________________
John McDonald
RETIRED |
|
| Back to top |
|
|
|
|
