Re: How are CPU Speed and Overall CPW Related?

["Alexei Pytel" <pytel@xxxxxxxxxx>]

I can suggest three possible explanations.
One is pretty straightforward - you may be mistaken in relation to cache
sizes.

Another is also quite simple.
RPG program always uses some runtime support - string manipulation BIFs
etc.
So if you have different OS releases, or PTF levels, then RPG program may
have different performance - due to software, not hardware differences. In
the same vein, different levels of compilers may produce different code
quality.

The last one is more far-fetched.
When you create a program, it is optimized for some target CPU (which is
different for different releases).
You then save and restore the program to a different box with different CPU
chips.
RISC processors are very sensitive to pipeline filling. Different chips
have different pipeline characteristics.
So the program highly optimized for Apache will not necessarily run
optimally on a newer chip.
I am not sure how to test it - try CHGPGM with forced program recreation.

    Alexei Pytel



                                                                                
                   
                    "Nathan M. Andelin"                                         
                   
                    <nathanma@haaga.com       To:     <MIDRANGE-L@midrange.com> 
                   
                    >                         cc:                               
                   
                    Sent by:                  Subject:     Re: How are CPU 
Speed and Overall CPW   
                    owner-midrange-l@mi        Related?                         
                   
                    drange.com                                                  
                   
                                                                                
                   
                                                                                
                   
                    05/01/2001 01:53 PM                                         
                   
                    Please respond to                                           
                   
                    MIDRANGE-L                                                  
                   
                                                                                
                   
                                                                                
                   



It's great to get a response from you, Alexei.  I think we all appreciate
your insider insight.  I won't bother adding more memory now, because at
this point I'm interested in raw CPU speed.

I'm aware that CPW is a measure pertaining to a mixed workload under load
testing and is affected by overall system design, but I still question
AS/400 CPU performance.

I wondered about a CPU governor for several reasons.  The program we were
benchmarking was entirely CPU bound.  The program simply repeated a few
String functions, in memory, maybe 100,000 times within an RPG do while
loop.  Both machines had about the same RAM.  Neither had L2 Cache.  I'm
talking about a model 170-2160 and a model 170-2290.  Yet the 100 Mhz
machine executed the code faster than the 200 Mhz machine.  If I were
running the same logic on a pair of Intel boxes, identical in every way
except Mhz, then I'm fairly certain the higher Mhz machine would be faster.

What makes the AS/400 CPU different, in this limited scenario?

Thanks,

Nathan.


> From: "Alexei Pytel" <pytel@us.ibm.com>
> Subject: Re: How are CPU Speed and Overall CPW Related?
>
> I see that now you suspect that IBM is putting a horde of governors all
> over the place.
> This is not the case.
>
> If your program uses little memory - it is not memory constrained - so
> adding more memory will not change anything for it.
> But if it requires lots of memory, and real memory is not big enough, and
> parts of a program or data get paged in and out - then your application
is
> memory constrained. Adding more memory will reduce or eliminate paging
> overhead and make your program run faster.
>
> Same logic applies to disk constrained - 100 disk drives can perform more
> disk accesses per second than 10 disk drives.
> So if your application needs 1000000 disk accesses to complete, then on a
> 100 drive system it will finish faster than on 10 drive system.
>
> That simple. And no need for governors...
>
>
> Best regards
>     Alexei Pytel
> System Performance III
> Dept XQK/006-2    Rochester, MN
> (507) 253- 2867 or T/L 553-2867
> Internet:  pytel@us.ibm.com  VM mail: IBMUSM07(PYTEL)


+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to
MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator:
david@midrange.com
+---




+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---