Re: how is system geared down? -- MIDRANGE-L


OK, last try...
Last week I tried to show that "system performance" is a misunderstood
concept.
Now let's have a look at "CPU performance". CPU is not an isolated
component. It's complex entity in itself, which consists of lots of
circuitry inside, L1 cache inside, L2 cache outside, main storage outside,
interconnection between registers and caches of different level, between
caches and memory etc etc
Next, different code has different characteristics - locality of reference,
working set size, pipeline affinity etc etc.
Next. don't forget also that you do not write code in chip instruction set.
Before your source code becomes executable it is worked on by compiler and
translator. Modern RISC chips rely very heavily on compiler technology -
to fill pipeline, to schedule instructions etc etc.

Raw chip Mhz rating tells you nothing - if it were, then 1Ghz Pentium would
be the king of processors once and for all - simpy because it has highest
clock rate.

Next, do not think that CPU chip with a certain Mhz clock rate should work
with the same speed in all circumstances with any code - this is a wrong
assumption.

Apache and Northstar are based on the same architecture, but they are
inherently different processors - they have different pipelines, different
instruction scheduling, different this, different that...

What I am arguing against, is your sweeping statement, that 200Mhz chip
should run twice as fast as 100Mhz chip and if this does not happen, then
something dark is under way.

Should your test run twice as fast on 200Mhz NorthStar then on 100Mhz
Apache ? - I don't know. Maybe yes, maybe not.
Your observations are interesting - it would be interesting to speculate
why this could happen. As a matter of fact, I offered you some suggestions
last week...
Nothing in your observations is an evidence of tampering with CPU.

    Alexei Pytel



                                                                                
                   
                    "Nathan M. Andelin"                                         
                   
                    <nathanma@haaga.com       To:     <MIDRANGE-L@midrange.com> 
                   
                    >                         cc:                               
                   
                    Sent by:                  Subject:     Re: how is system 
geared down?          
                    owner-midrange-l@mi                                         
                   
                    drange.com                                                  
                   
                                                                                
                   
                                                                                
                   
                    05/09/2001 04:11 PM                                         
                   
                    Please respond to                                           
                   
                    MIDRANGE-L                                                  
                   
                                                                                
                   
                                                                                
                   



> From: "Alexei Pytel" <pytel@us.ibm.com>

>You can use the same CPU chip and have
> systems with different performance ...

That's easy to understand.  "Systems" have many points where bottlenecks
can
occur.  But the real puzzle is why a 200 Mhz CPU offers the same throughput
as a 50 Mhz CPU, for example.  Again, it's not "system" throughput that's
puzzling some of us, but "CPU" throughput.  An explanation of how memory,
cache, paging, disk I/O, etc. might affect system throughput only clouds
the
issue.  Customers can add memory, add disks, perform tuning to adjust
paging, and design or configure software to achieve optimum system
performance.  But a customer apparently can't do anything to change CPU
speed.  The real puzzle is with CPU performance, not system performance.

Thanks,

Nathan.


+---
| 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
+---