Other than MIPS...



Ask queries about other IBM Tools like Tivoli, COBTEST, Fault Analyzer, z/OS File Manager, Workload Simulator, APA, SCLM, Merge & Migration Tools etc...

Other than MIPS...

Postby edotm » Wed May 27, 2009 8:12 pm

Hi all,

First, I'm glad I found a such place like this :D ! Sencond, I would like to get some opinions here, as this is my first time working with SMF/RMF :( .

As we all know, no existing MIPS table can adequately quantify the capacity of one System z processor against another. Every documentation remind us to "Never use a MIPS table for capacity planning purposes!"

Right, so I came with some numbers I found in the RMF's “Workload Activity Report”, specifically the “Workload Group and Service Class Period Report”. The “Processor Engine requirement” – or how much an given application needs from a processor (%) in order to work. Well, at least this is my understanding... :?:

My understanding is that in this report, we have relevant values under CP field in APPL% session, which shows the number of processor engines required to drive the work. In my example here (covering Broker and DB2 classes), the relevant active transactions in the system (field AVG under TRANSACTIONS) require the given % of a processor engine (one CP) to support their work.

W O R K L O A D   A C T I V I T Y
REPORT BY: POLICY=NORMAL   REPORT CLASS=BROKER
--TIME--    -TRANSACTIONS-   ---APPL %---      ----STORAGE----
11.10.00    AVG   6.00      CP    1.62      AVG   85299.88
11.15.00    AVG   6.00      CP   21.17      AVG   85442.83
11.20.00    AVG   6.00      CP    4.34      AVG   85987.75
11.25.00    AVG   6.00      CP   20.32      AVG   85855.58
11.30.00    AVG   6.00      CP   21.05      AVG   85651.33
11.35.00    AVG   6.00      CP    0.25      AVG   85990.53

REPORT BY: POLICY=NORMAL   REPORT CLASS=DB2
--TIME--   -TRANSACTIONS-   ---APPL %---      ----STORAGE----
11.10.00   AVG    9.07      CP    0.34     AVG   14545.78
11.15.00   AVG   10.00      CP    2.76     AVG   13262.00
11.20.00   AVG   10.00      CP    2.72     AVG   13262.43
11.25.00   AVG   10.00      CP    2.66     AVG   13262.32
11.30.00   AVG    9.34      CP    0.67     AVG   14125.01
11.35.00   AVG    9.16      CP    0.15     AVG   14379.83


So I think I can use the following formula:
SUM(APPL CP) ÷ SUM(AVG TRANSACTIONS)= Avg CPU % / Transaction

which in my example is
6.50 ÷ 9.08 = 0.716% of 1 CPU per transaction

The Total CPU Runtime is not considered in these numbers, but only the CPU Room Time; any waiting time (by resources, contention or any other reason) is not considered. Saying this, the same message can take 1 minute in one machine but 15 seconds in another and still have the same rates for its CPU % per transaction.

My conclusion is that using this methodology, it doesn't matter the machine's model, how powerful is the processor, nor how many processors it has: the numbers will show the percentage needs of 1 single processor to run 1 single transaction in the system. In this way, I'm able to compare the numbers in ANY machine, with their own parameters.

Am I going crazy? :shock:

-----
Cheers,
e.m
User avatar
edotm
 
Posts: 1
Joined: Wed May 27, 2009 7:35 pm
Has thanked: 0 time
Been thanked: 0 time

Return to Other IBM Tools