View previous topic :: View next topic
|
Author |
Message |
Jagan Sachidanandam
New User
Joined: 11 Aug 2008 Posts: 14 Location: India
|
|
|
|
Reducing MIPS of Mainframe CPU
Hi
We are trying to calculate the MIPS usage of a CICS transaction with the below formula,
MIPS Usage = CPU time * Speed of the instruction processor
Where,
Speed of the instruction processor is 564.3 since we use z10.
Can you please confirm if this method is correct? Or suggest any other options.
Thanks in Advance |
|
Back to top |
|
|
enrico-sorichetti
Superior Member
Joined: 14 Mar 2007 Posts: 10873 Location: italy
|
|
|
|
Quote: |
Can you please confirm if this method is correct? Or suggest any other options. |
it all depends on Your expectations, but as a general point it is just plain wrong
MIPS ==> MEANINGLESS INDICATOR of PROCESSOR SPEED
if Your organization is trying to do some Capacity Planning/Verifcation there are better indicators available |
|
Back to top |
|
|
Jagan Sachidanandam
New User
Joined: 11 Aug 2008 Posts: 14 Location: India
|
|
|
|
Thanks Enrico.
Yes, we use the same expansion for MIPS . However I am trying to use this formula this as we have an requirement specified to calculate MIPS usage. |
|
Back to top |
|
|
Robert Sample
Global Moderator
Joined: 06 Jun 2008 Posts: 8697 Location: Dubuque, Iowa, USA
|
|
|
|
Quote: |
We are trying to calculate the MIPS usage of a CICS transaction with the below formula,
MIPS Usage = CPU time * Speed of the instruction processor
Where,
Speed of the instruction processor is 564.3 since we use z10. |
Questions come to mind:
Where did the 564.3 come from?
How are you measuring CPU time for a CICS transaction?
Are you considering SRB and TCB time as CPU time?
Is the LPAR defined to use multiple processors?
Do you have MXG or MICS available?
Are you calculating MIPS usage for migration impact or chargeback -- and what about system overhead? |
|
Back to top |
|
|
Jagan Sachidanandam
New User
Joined: 11 Aug 2008 Posts: 14 Location: India
|
|
|
|
Hi Robert,
Thanks for the queries. The answers are
1. 564.3 is the value provided by our system programmers for one CPU
2. We measure CPU time using TMONCICS.
3. CPU time provided in TMON has SRB and TCB time in it. I am not sure about this though. I will check with the system programmers
4. The LPAR has got only one CPU and the MIPS is Capped
5. I have not heard about these term MXG and MICS. I will do a search for this.
6. We have asked to calculate MIPS usage, because of a new application being introduced into the system. |
|
Back to top |
|
|
Robert Sample
Global Moderator
Joined: 06 Jun 2008 Posts: 8697 Location: Dubuque, Iowa, USA
|
|
|
|
MXG and MICS are vendor products that process SMF data and can report on CICS transaction details -- probably what your TMONCICS is doing.
How many transactions are you using in your measurement? We have transactions that can run from a few thousandths of a second of CPU time up to minutes (depending on what the person is doing) so you need to make sure you've got a good cross-section of the variability of the transaction.
Your number needs to include TCB and SRB time since that's a better picture of the actual time the transaction took -- measuring just the task (TCB) time may understate how much actual CPU time is being used.
Just out of curiosity, why isn't the NEW application giving this data since the existing workload presumably is better understood than the additional workload? |
|
Back to top |
|
|
Jagan Sachidanandam
New User
Joined: 11 Aug 2008 Posts: 14 Location: India
|
|
|
|
Hi Robert,
The transactions originate from an external web interface. We have limited the number of concurrent transactions to 14 (business decision) using the MAXSSLTCB parameter. We measure the CPU time for 500 transaction and verify the response times for each of them. And we do have the think time included and variability of the transactons will be used. Yes, i confirmed with system programmers they mentioned the CPU time measured is TCB time + SRB time. We have integrated the external interface with the CICS and we need to understand if N users use the interface how is the response time affected in the existing CICS system and the external system.
Thanks |
|
Back to top |
|
|
Robert Sample
Global Moderator
Joined: 06 Jun 2008 Posts: 8697 Location: Dubuque, Iowa, USA
|
|
|
|
It sounds like everything that needs to be considered has been -- it is still a meaningless number but at least it is a sensible meaningless number.
I can't believe I just said that. |
|
Back to top |
|
|
Jagan Sachidanandam
New User
Joined: 11 Aug 2008 Posts: 14 Location: India
|
|
|
|
Hi,
Just out of curiosity, I am always hearing that there are other parameters to measure while doing performance improvement. Can you please let me know few other parameters that will be useful? We measure response time, I/O time and I/O count apart from CPU time.
Thanks |
|
Back to top |
|
|
MBabu
Active User
Joined: 03 Aug 2008 Posts: 400 Location: Mumbai
|
|
|
|
in general, anything considered the "working set" of a program can be looked at. That would be CPU time, memory use, files, common system resources, wait times, ENQs, transmission times, etc. I don't know how these relate to CICS systems since CICS manages most of these resources itself.
As for performance improvement, the biggest thing to look for is good algorithms. Then look at the minutia. Reducing memory use is always a good thing if it isn't done at the expense of other resources (eg, you wouldn't want to swap small amounts of data to a file just to reduce the size of an array) because less memory tends to mean less paging and that is good for everyone. But again, your transactions might be sharing some memory (can CICS do that?) or have other interesting design points. For example, it may be possible to use different types of programs interpreting XML (cobol parses, your own parsers, CICS interpretive parsers, RD/z generated parsers, etc). I'll have to let someone who knows CICS answer questions about resource attributes, file tuning etc affect CICS. I admit I only answered because I felt like typing |
|
Back to top |
|
|
Jagan Sachidanandam
New User
Joined: 11 Aug 2008 Posts: 14 Location: India
|
|
|
|
Thanks Rob. We did check the design of Cobol parser programs. We proposed the business to have the XML parsing done outside mainframe, like say CICS transaction server, but you know business they didnt listen to us for some reason. |
|
Back to top |
|
|
|