Page 1 of 1

Extracting data from SMF 70

PostPosted: Thu Mar 11, 2010 12:05 pm
by avik1983
I want to derive Latent Demand Estimate for my mainframe system.For getting that I need to know some parameters (SMF70ITT,SMF70INT,SMF70RTT,NUMCP,DISPCP etc) from SMF70 records.
As I couldn't read the values directly from SMF70 record,can anyone help regarding this.If someone can provide a program for extracting those values from SMF70 Dump ,it will be great.

Re: Extracting data from SMF 70

PostPosted: Thu Mar 11, 2010 6:32 pm
by Robert Sample
What language are you planning on using? I typically use SAS to access the SMF data since it handles the bit variables so well, but not all sites have SAS installed.

Re: Extracting data from SMF 70

PostPosted: Thu Mar 11, 2010 9:15 pm
by Bill Dennis
How about running standard RMF PostProcessor CPU reports and look at OUT READY queue counts?

Re: Extracting data from SMF 70

PostPosted: Fri Mar 19, 2010 11:07 am
by avik1983
I found this calculation from one whitepaper.Is this correct? If so then what about when I have ziiP/ZaaP in my shop.How to calculate the CPU utilization?

Example of Latent Demand Estimation

If one measures an 8-way processor in a
measurement interval and observes:
CPU = 89.62%
IN = 242.26
INRDY = 31.26
The following is computed:
U = 89.62 / 100 = 0.8962
DISPCP = (89.62 / 100) x 8 = 7.17
WAITCP = 31.26 – 7.17 = 24.09
IN* = 242.26 – 24.09 = 218.17
U* = (241.26 / 218.17) x 0.8962 = 0.9910
Thus, the estimated latent demand is U* - U = 0.9901
– 0.8962 = 0.0939 = 9.39%. This represents the
increase in CPU utilization that would result if all of the
delay were removed at the CPU resource.

Re: Extracting data from SMF 70

PostPosted: Fri Mar 19, 2010 4:20 pm
by Robert Sample
CPU utilization requires rethinking when zIIP and zAAP engines are involved. Google is your friend: cpu utilization ziip returns about 4,980 hits and some of them, even on the first page of results, talk about the impact of the new engines on CPU measurement. But basically, measuring CPU utilization becomes more complicated since work can be zIIP-eligible, zIIP-qualified, GP (general processor), and overflow (zIIP-qualified work that ran on the GP because the zIIP engine couldn't get it done).

You would be wise to look at CMG's web site (http://www.cmg.org) as well since I'm sure they have had some interesting discussions on measurement of CPU with the new engines.