| Author |
Message
|
| lancelotlinc |
 Posted: Mon Jan 31, 2011 5:25 am Post subject: WMB runtime performance on z/OS compared to Power7 |
 |
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
Hi.
When performing some analysis of WMB processing capacity on z/OS, I noticed that due to a system software problem in the z/OS operating system, it takes more than twice as much CPU time to process half as much data in z/OS compared to AIX on Power7 per the following documentation:
AIX https://www-304.ibm.com/support/docview.wss?uid=swg24025868
z/OS ftp://public.dhe.ibm.com/software/integration/support/supportpacs/individual/ip14.pdf page 49
| Quote: |
| This drop in throughput for 8 instances when compared with 4 instances is being investigated and a fix will be shipped in due course. Contact IBM service for the latest status on this problem. |
Has anyone heard of the fix being deployed to correct this? What specifically is involved in this fix?
Also, to be more fair to z/OS, I'm looking for and not finding an update to the performance report for WMB version 7 on z/OS.
TIA for any information you can provide.
Lance
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
 |
| mqjeff |
| Posted: Mon Jan 31, 2011 5:38 am Post subject: |
|
|
Grand Master
Joined: 25 Jun 2008
Posts: 17447
|
Likely the performance report was run against broker 7.0.0.0.
I'd start by looking at the PTFS that are included in Broker 7.0.0.2. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Mon Jan 31, 2011 6:41 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
I notice in IP15 that the reference message sizes are 4k and in IP14 the reference message sizes are 1k. In both cases, it takes twice as much CPU time to process half as many messages as when looking at AIX Power7 CPU time.
It seems that there is extra special effort to avoid the direct comparison between the platforms. This seems to be intentional so as not to foment dissatisfaction with the legacy platform. Power7 seems to be able to deliver twice as much work in half the time.
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Mon Jan 31, 2011 7:05 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| It seems that there is extra special effort to avoid the direct comparison between the platforms. This seems to be intentional so as not to foment dissatisfaction with the legacy platform. |
That's quite an assertion. Please cite your sources.
| lancelotlinc wrote: |
| Power7 seems to be able to deliver twice as much work in half the time. |
I'm interested in your sources, and your soon-to-be provided disertation that the extra throughput provided such benefits that everyone who's chosen or chosen to remain with z/OS will abandon it for distributed AIX en masse.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Mon Jan 31, 2011 7:11 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
Sir Vitor,
I have extra special appreciation for you. There are other reasons to stay on z/OS besides performance. One may be migration cost.
The sources were cited above: IP14, IP15, IPL6, IP6N, IP08, IP76.
What is missing, and I have not been able to locate, is a WMB v7 performance report for z/OS.
Lance
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Mon Jan 31, 2011 7:24 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| The sources were cited above: IP14, IP15, IPL6, IP6N, IP08, IP76. |
Not the assertion regarding performance, but the assertion that the missing report is deliberately absent.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Mon Jan 31, 2011 7:28 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
|
| Back to top |
|
|
| mqjeff |
| Posted: Mon Jan 31, 2011 7:29 am Post subject: |
|
|
Grand Master
Joined: 25 Jun 2008
Posts: 17447
|
| lancelotlinc wrote: |
| Maybe I'm not looking in the right place. Can you point me to the report? |
He is not disagreeing with you that the report is missing.
He is disagreeing with your speculations as to why the report is missing. |
|
| Back to top |
|
|
| Vitor |
| Posted: Mon Jan 31, 2011 7:31 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| mqjeff wrote: |
| He is disagreeing with your speculations as to why the report is missing. |
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Mon Jan 31, 2011 7:40 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
We could all be happy if we found the report.
Nevertheless, to appease my friend, I will elaborate on some thoughts. The problem any Java program has on z/OS is the lack of VLCMBs (Very Large Contiguous Memory Blocks). This impacts the ability of the JVM to deliver requested memory blocks without undue frequent calls to GC process to free up bits of memory. This is the well known and very well documented memory fragmentation slowness that affects all Java computing platforms, including z/OS, Unix and Windows.
The Power7 architecture has done it equisitely well in that the designers of the architecture have seen fit to grant eight terabytes of memory address space per core, with access to a global memory pool of 2 petabytes. This creates a huge pool of memory from which VLCMBs can be easily and quickly allocated without the memory fragmentation slowness previously described.
In IP14 (WMB 6.0 on z/OS), the performance tests measured 1k message sizes against multiple execution groups, resulting in CPU times of 2.3 millis. In IP15 (WMB 6.1 on z/OS), the same performance tests were measured against 4k messages. Sort of sidestepping the defacto standard in all the other performance reports of 2k message sizes. What is the reason for this?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Mon Jan 31, 2011 7:45 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| We could all be happy if we found the report. |
I didn't ask you for further proof or elaboration on the performance (good or bad), the architecture or any other technical points.
I asked you to support and defend your claim that the report has intentionally not been released.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Mon Jan 31, 2011 7:51 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
My assertion involves the lack of direct comparison of performance between platforms. I am searching for WMB v7 on z/OS performance report.
While I believe there to be more than one gunman in November, 1963, I assume that all such documentation evidence has been redacted or destroyed. Else, why lock the documents up for 99 years?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Mon Jan 31, 2011 8:00 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| My assertion involves the lack of direct comparison of performance between platforms. I am searching for WMB v7 on z/OS performance report. |
So you refute the possible explaination that it's simply not been produced or released due to time or other constraints?
| lancelotlinc wrote: |
| While I believe there to be more than one gunman in November, 1963, I assume that all such documentation evidence has been redacted or destroyed. Else, why lock the documents up for 99 years? |
This is a technical forum, not a conspiracy theory web site. Do not throw that sort of thing round without evidence; your paranoia is not evidence. If you want these details, raise a PMR and ask for it. I'll consider IBM's refusal to provide it or a publication date evidence. In the meantime, keep your personal views off the board.
I accept you now consider me part of the greater IBM z/OS conspiracy and probably related to the second gunman.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| mqjeff |
| Posted: Mon Jan 31, 2011 8:24 am Post subject: |
|
|
Grand Master
Joined: 25 Jun 2008
Posts: 17447
|
On the subject of 1k vs. 4k messages.
Given the amount of time that passed between the two performance reports - several years - is it *unlikely* or *likely* that the average size of messages in use by actual customers in actual production would have changed?
If it is *likely* that the size of actual messages in actual production use by actual customers would have *increased* since the last performance report, then what is the correct thing to do?
Produce another performance report that uses the same size as the old one, that is then NOT representative of expected customer scenarios?
Or produce another performance report that is MORE representative of actual customer scenarios, even if it then means that there is LESS comparison with the previous one?
Second set of questions:
You have shown that a known performance issue with Broker v7 on zOS exists. You have shown that IBM acknowledges this issue.
Do you believe that is it MORE appropriate or LESS appropriate to produce a performance report that includes this known defect, or to delay producing a performance report until this known defect is resolved? |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Mon Jan 31, 2011 8:50 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
Wow, must have been a rockin weekend. People are up-tight.
Take a chill. Relax. Life is good.
All these are viable possibilities. And it is appropriate not to draw direct conclusions between platforms as the comparisons are not apples to apples, ever, even if the message sizes were the same. Still, it is a good exercise to see the results when then same use cases with the same message sizes were used.
I like z/OS. I am very supportive of the whole idea behind it, and it has endured the test of time. Many companies have built so heavily on z/OS that their entire business depends on that platform and will likely never be able to be moved off.
I like Power7 even more because of the monumental and game-changing advances that the expert designers have built into the vision. Truly amazing.
The opportunity I see at this site is moving a portion of the processing to Power7, where it makes good business sense and economic sense to do so. Thus my initial query. Perhaps there is a good reason for the delay in the z/OS report for WMB v7. Or maybe I just MISSED it and someone may know WHERE it is?
And with even more hope, maybe the WMB v7 on z/OS report will use 2k messages so we can see the comparison of CPU time per message between z/OS, AIX, HP-UX, and Windows.
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
|
|
