[prev in list] [next in list] [prev in thread] [next in thread] 

List:       munt
Subject:    Re[2]: WinBench comparison of NTRIGUE and WinCenter Pro
From:       "zdravko.podolski () isinc ! insignia ! com" <zdravko ! podolski () isinc ! insignia ! com>
Date:       1996-07-31 10:33:07
[Download RAW message or body]

Reinhard asks some excellent questions. 

Of course I already believe NTRIGUE is the better product, just like Doug Klein
believes his is better  ;-), but that is a different issue. 

The main reason why cpu utilization is not 100% is that the benchmark exercises
other computer subsystems as it runs and therefore the CPU is not fully loaded.
For the single user case, NTRIGUE uses less cpu load and gets faster results. If
we were to use a pure cpu benchmark then we would be benchmarking the hardware
speed which is not a function of NTRIGUE or WinCenter, but of the hardware
design.

The fact that a "normal" application does not exercise the cpu fully is one of
the crucial reasons why it is possible to have a multi user application server
where the users all seem to get more than simple division would indicate. While
one is waiting for disk IO to complete, another gets the cpu and a third gets
the network, etc. You add all that together and the total is greater than the
sum of its parts. The less cpu a single user benchmark consumes, the more users
the cpu will support. Of course there are other potential bottlenecks in a
system, as Reinhard suggests.

By the way, even if you run WinBench on a vanilla NT system with a standard VGA
console, the CPU will be idle a significant proportion of the time, since the
video frame buffer is not as fast as the CPU and introduces wait states. 

We have other benchmarks that measure what happens as you add load to the
system. I'd be happy to discuss benchmarking quite apart from product oriented
discussions, as I believe it is in everyone's interest for any measurements to
be done in a manner that is meaningful, repeatable and comparable.

Zdravko
_______________________________________________________________________________
Subject: Re: WinBench comparison of NTRIGUE and WinCenter Pro
From:    reini@x-tech.co.at (Reinhard Travnicek) at MVSMTPLINK
Date:    31/07/96  15:55

On Jul 30,  4:31pm, Zdravko wrote:
>
>                         Insignia                NCD
>      system             Perf    CPU Util        Perf    CPU Util
>
>      IBM PPC604-132     6.1     35.0%           5.3     50.0%
>      SGI Indy R4600-132 16.7    50.4%           13.5    65.9%
>      Sparc 20           11.1    42.3%           9.4     56.0%
>      PC Pentium-100, NT 5.3     25.2%           3.6     35.0%
>         (Hummingbird 5.0)
>      NCD HMX            7.47    28.1%           7.34    42.0%
>      HP Envizex         6.50    29.1%           6.25    41.9%
>
Zdravko,

there is no doubt, that this benchmark shows Insignia has the faster product.
In our benchmark driven world this will lead to the opinion that Insignias
product is the better.

What makes me thinking (and that's true for both products), why the hell is
a application not running at 100% CPU load ?
If I have a "super" computer (like the Pentium Pro 200 you mentioned doubtless
is), why is a graphics benchmark (as the only application on the system) not
running at 100% CPU load ?

There is a nice little section in the Windows NT resource kit talking about
finding bottle necks in application.

Some ideas from there:

Is the system doing context switches, or is the I/O (ethernet) the bottleneck,
is the X driver sending to much packages ????

Many questions ? Are there any good answers ?

regards

Reinhard



-- 
Reinhard Travnicek                     email: reinhard@x-tech.co.at
Tech. Manager                      snailmail: X-tech HandeslgesmbH
                                              Peter Strassergasse 35
Tel.: (43) 2287 3928                          A-2231 Strasshof
Fax.: (43) 2287 3928 21                         Austria


--------------<< "multiuser-nt@ncd.com" FOOTER MESSAGE >>-------------------

To Unsubscribe:  Email ncd-majordomo@ncd.com with "unsubscribe multiuser-nt".
Archive of List: Use "index" and "get" commands to ncd-majordomo@ncd.com
NCD reserves the right to restrict access to this list.

[prev in list] [next in list] [prev in thread] [next in thread]