I used it about 10 years ago.  We only looked at hot hot spots. Small hot
spots close to the background level, could change from day to day depending
on other usage.  For example are the instructions/data cached in the local
processor?  Changing the amount of optimization of the C code made a big
difference.  Sometimes highly optimised code was slower than medium
optimised, because it optimised the whole program, whereas the hot code was
only about 5% of the total.  Of course upgrade the processor, and
everything can change, bigger cache, bigger page size, TLB etc.

I remember going to a customer to resolve a performance problem who was
going live in under 2 weeks.  Using APA...
1)  The top usage in the top transaction was 80% in  "printf".   They still
had debug code running.
2)  Rather than use a variable in a dynamic SQL statement  such as "select
from table where user=:userid", they had "select from table where
user='COLIN'", and "select from table where user='PAICE', so each of these
statements were unique, and could not be cached.
APA showed me these in the first hour (it made me look great). Once fixed,
the CPU dropped from 4 engines down to 1 engine for the same workload.
When I said they were going live in under 2 weeks; every one used the
password "qw", and I could logon to the super user using qw!

Colin





On Tue, 13 Apr 2021 at 16:56, Phil Smith III <li...@akphs.com> wrote:

> Colin Paice wrote:
>
> >I dont think it matters which machine you run on, you just run for a
> longer
>
> >time, and get more samples that way.
>
>
>
> There was, IIRC, also a maximum time for the sampling. What we wound up
> with
> was insufficient; as I noted, it was a while ago. Perhaps we missed
> something.
>
>
> ----------------------------------------------------------------------
> For IBM-MAIN subscribe / signoff / archive access instructions,
> send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
>

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN

Reply via email to