Author |
Message
|
KoGor |
Posted: Fri Jul 19, 2013 2:49 am Post subject: Limits of one execution group |
|
|
Voyager
Joined: 09 Nov 2005 Posts: 81 Location: Moscow,Russia.
|
Hi all!
I have to decide what version of message broker - standard or advanced we are going to use in our new project. We are going to use WMB as HTTP back-end and it's expected that it would be about 3000 concurrent HTTP client session to the MessageFlow. A MessageFlow should parse, put to DB2 incoming HTTP message, then put it to WMQ and reply to HTTP client the result of transaction. That's all.
As IBM spec says standard edition can have only one EG. But what are the limits of one execution group? I only found that it can run only 256 threads and that one EG is bound to only one CPU. Is this is really true? I have not found a proof in WMB documentation.
Also I've found that default setting for HTTP input node is only 255 concurrent sessions. But there is no information about maximum HTTP session that one EG or MessageFlow can handle.
So I'm really stuck with the question what version of WMB is enough for our project. Any advice is appreciated! |
|
Back to top |
|
 |
lancelotlinc |
Posted: Fri Jul 19, 2013 2:52 am Post subject: |
|
|
 Jedi Knight
Joined: 22 Mar 2010 Posts: 4941 Location: Bloomington, IL USA
|
|
Back to top |
|
 |
KoGor |
Posted: Fri Jul 19, 2013 3:02 am Post subject: |
|
|
Voyager
Joined: 09 Nov 2005 Posts: 81 Location: Moscow,Russia.
|
I've seen this page. As I wrote I know the difference between WMB versions. The question is about one EG limits. |
|
Back to top |
|
 |
lancelotlinc |
Posted: Fri Jul 19, 2013 3:07 am Post subject: |
|
|
 Jedi Knight
Joined: 22 Mar 2010 Posts: 4941 Location: Bloomington, IL USA
|
KoGor wrote: |
I've seen this page. As I wrote I know the difference between WMB versions. The question is about one EG limits. |
An EG is a unique process. As performance is entirely dependent on your message flows, one who has no visibility on same cannot give you guidance.
As a general reference, you can read the performance reports for WMB.
ftp://public.dhe.ibm.com/software/integration/support/supportpacs/individual/ip6p.pdf
What version of WMB are you using (version numbers have four digits)? What OS are you using? _________________ http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
Back to top |
|
 |
KoGor |
Posted: Fri Jul 19, 2013 3:19 am Post subject: |
|
|
Voyager
Joined: 09 Nov 2005 Posts: 81 Location: Moscow,Russia.
|
I also have seen this report But there is no answer to my question. Now I need only to understand what are limits of one EG (actually it doesn't matter advanced or standard edition, I think that EG doesn't have any difference between editions)
Now we do only some research for upcoming project. There is no any MessageFlow , just the request for proposal. So we have to estimate and offer to a customer info about hardware and software versions to help them plan the budget on the next year. We are definitely going to use AIX 7.1 and I think we are going to WMB 8 or 9 (integration bus) if there would be a couple of fix packs at that time  |
|
Back to top |
|
 |
lancelotlinc |
Posted: Fri Jul 19, 2013 3:31 am Post subject: |
|
|
 Jedi Knight
Joined: 22 Mar 2010 Posts: 4941 Location: Bloomington, IL USA
|
Page 64 of the performance report indicates relative limits for one EG at around 1,785 payloads per second:
Quote: |
Number of
Execution
Groups
Persistent
Message
Rate
(Msgs/Sec)
% CPU
Busy
CPU
ms/msg
1 NONE 1785.2 10.1 0.5
2 NONE 3245.9 22.2 0.6
3 NONE 4801.8 32.8 0.6
4 NONE 5824.7 41.7 0.6
5 NONE 6513.2 49.0 0.6
6 NONE 7368.2 56.7 0.6
7 NONE 8001.2 63.8 0.6
8 NONE 8362.4 68.3 0.7
9 NONE 8477.7 70.4 0.7
10 NONE 8624.0 72.0 0.7
11 NONE 8744.9 74.9 0.7
12 NONE 8982.5 77.9 0.7
13 NONE 9196.3 80.7 0.7
14 NONE 9461.6 82.8 0.7
15 NONE 9822.1 83.6 0.7
16 NONE 9985.7 85.4 0.7
17 NONE 10312.5 86.8 0.7
18 NONE 10381.2 88.5 0.7
19 NONE 10626.6 90.0 0.7
20 NONE 10799.0 91.4 0.7
21 NONE 10953.4 92.6 0.7
22 NONE 10992.8 93.9 0.7
23 NONE 11122.6 95.3 0.7 |
It doesn't really matter what the transport or input node is, because all the input nodes do pretty much the same thing. In fact, SOAPInput nodes from the runtime listener actually use MQs in-between. _________________ http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
Back to top |
|
 |
KoGor |
Posted: Fri Jul 19, 2013 5:24 am Post subject: |
|
|
Voyager
Joined: 09 Nov 2005 Posts: 81 Location: Moscow,Russia.
|
But still didn't catch what is the difference between running multiple message flow instances on one EG and several EGs running same message flow to achieve the same result. What are hardware limits (CPU, memory and etc ) should one EG reach so I have to deploy new message flow to a new EG. |
|
Back to top |
|
 |
lancelotlinc |
Posted: Fri Jul 19, 2013 5:30 am Post subject: |
|
|
 Jedi Knight
Joined: 22 Mar 2010 Posts: 4941 Location: Bloomington, IL USA
|
KoGor wrote: |
But still didn't catch what is the difference between running multiple message flow instances on one EG and several EGs running same message flow to achieve the same result. What are hardware limits (CPU, memory and etc ) should one EG reach so I have to deploy new message flow to a new EG. |
No one can tell you this because your message flow has different performance characteristics than someone else's message flow. And since your dealing with hypotheticals, what difference does it make if its 10 instances and one EG or three instances and four EGs?
Why don't you download the trial and see for yourself? Nothing like hard, cold numbers to prove or disprove your theory. _________________ http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
Back to top |
|
 |
KoGor |
Posted: Fri Jul 19, 2013 6:04 am Post subject: |
|
|
Voyager
Joined: 09 Nov 2005 Posts: 81 Location: Moscow,Russia.
|
So as I understand IBM doesn't have any documentation that would describe one EG limits (for example running only on 1 CPU, 256 threads limit, 4GB memory limit and so on) Should I do conclusion that there is no actually best practices or recommendations for Message flow and EG usage and actually there is no big difference if you use multiple message flow instances or several EG?
P.S. I forget to answer what differences does it take for me. Actually there is a big difference in price between standard and advanced edition os WMB. I think that's a big one for a customer!  |
|
Back to top |
|
 |
lancelotlinc |
Posted: Fri Jul 19, 2013 6:19 am Post subject: |
|
|
 Jedi Knight
Joined: 22 Mar 2010 Posts: 4941 Location: Bloomington, IL USA
|
KoGor wrote: |
So as I understand IBM doesn't have any documentation that would describe one EG limits (for example running only on 1 CPU, 256 threads limit, 4GB memory limit and so on) |
There is documentation in the form of performance reports on many platforms. You can read these reports and understand what expectations you can have with regard to your own performance. No documentation can ever exist that will precisely call out what YOUR performance will be. How could it?
KoGor wrote: |
Should I do conclusion that there is no actually best practices or recommendations for Message flow and EG usage and actually there is no big difference if you use multiple message flow instances or several EG? |
There is no such thing as 'best' practice when it comes to Message Broker. There are 'good' practices. There are 'not-so-good' practices. But there are no 'best' practices.
There are reasons you might run one EG and there are other reasons why you might run many EGs. Thats the 'art' of the state. Thats why WMB developers command multiple hundred dollar per hour bill rates.
KoGor wrote: |
P.S. I forget to answer what differences does it take for me. Actually there is a big difference in price between standard and advanced edition os WMB. I think that's a big one for a customer!  |
Start small and when you bump your head with no more development changes to make for inefficiencies' sake, then add the more expensive license. _________________ http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
Back to top |
|
 |
zpat |
Posted: Fri Jul 19, 2013 6:48 am Post subject: |
|
|
 Jedi Council
Joined: 19 May 2001 Posts: 5866 Location: UK
|
The low cost version is intended (in my opinion) for edge applications.
For your sort of major usage and critical function you should use the Enterprise version so that you can use the product to its full capability.
This is purely my view and I don't work for IBM. |
|
Back to top |
|
 |
Vitor |
Posted: Fri Jul 19, 2013 7:15 am Post subject: |
|
|
 Grand High Poobah
Joined: 11 Nov 2005 Posts: 26093 Location: Texas, USA
|
zpat wrote: |
The low cost version is intended (in my opinion) for edge applications. |
Or for any installation where you don't need all the bells and whistles. We're looking through the estate to where we can save a few $$ by using a lower level license on some servers, which will use MQ/WS/JSON to hand off to more expensively licensed machines where the application requires the functionality.
zpat wrote: |
This is purely my view and I don't work for IBM. |
Likewise. _________________ Honesty is the best policy.
Insanity is the best defence. |
|
Back to top |
|
 |
digoshc |
Posted: Fri Jul 19, 2013 8:47 am Post subject: |
|
|
Apprentice
Joined: 11 Jul 2013 Posts: 25
|
Kogor, I searched by 'execution group capacity' in the forum search and found some good threads about this. I hope can help you. |
|
Back to top |
|
 |
KoGor |
Posted: Thu Sep 26, 2013 2:38 am Post subject: |
|
|
Voyager
Joined: 09 Nov 2005 Posts: 81 Location: Moscow,Russia.
|
Finally I've got my answers. I have had a long discussion with IBMers about this question.
Finally we came to the conclusion that one EG is running as single OS process (I was interested in AIX) And every message flow from this EG are threads for this process. And according to this article:
http://www.ibm.com/developerworks/aix/library/au-aixprocesscontrol/
it seems that threads (read message flows) within one EG process can be bound to different CPU (cores).
So this is the answer I was searching for. Because, on the other case (the one process with all its threads are running only on one CPU) there is no use to assign to WMB's LPAR more that one or two CPU (the second CPU for MQ for example, but not for WMB), because they won't be used by WMB anyway. |
|
Back to top |
|
 |
lancelotlinc |
Posted: Thu Sep 26, 2013 3:08 am Post subject: |
|
|
 Jedi Knight
Joined: 22 Mar 2010 Posts: 4941 Location: Bloomington, IL USA
|
KoGor wrote: |
Finally I've got my answers. I have had a long discussion with IBMers about this question.
Finally we came to the conclusion that one EG is running as single OS process (I was interested in AIX) And every message flow from this EG are threads for this process. And according to this article:
http://www.ibm.com/developerworks/aix/library/au-aixprocesscontrol/
it seems that threads (read message flows) within one EG process can be bound to different CPU (cores).
So this is the answer I was searching for. Because, on the other case (the one process with all its threads are running only on one CPU) there is no use to assign to WMB's LPAR more that one or two CPU (the second CPU for MQ for example, but not for WMB), because they won't be used by WMB anyway. |
What is different about what you just posted and the discussion above ? It seems you are rehashing the same information. _________________ http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
Back to top |
|
 |
|