Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.java.programmer > #5836

Re: baseline performance test using java ...

Show key headers only | View raw

On 7/3/2011 10:53 AM, lbrt chx _ gemale kom wrote:

It would be better to reply to one article at a time, and do so in the
same thread, rather than starting a new thread with a mixture of replies
to different articles by different people.

...
>> In general, benchmarking, like any experiment, only makes sense if
>> you have some questions you want answered, and the measurements
>> will answer those questions. I'm not at all clear what your
>> questions are.
>
> The q is then. How much more (as a relative measure/percentage) does
> my logical stack taxes the underlying (hw + sw) baseline?

I still don't understand you. How do you define "taxes" and "the
underlying (hw + sw) baseline"?

>
>> "Performance of an algorithm" is a non-trivial concept ... matter
>> of analysis, not measurement.
>
> I am not sure if I got that one, but if we are talking here about the
> implementation of an algo then surely can we for instance check how
> much memory and time needs to run

Dropping the material you replaced with "..." tends to change the
meaning of the sentence. I do not think that all algorithm performance
work is a matter of analysis.

If you are interested in computer performance, and are not already
familiar with the concept of computational complexity, I suggest looking
into it. The Wikipedia page is a good starting point:
http://en.wikipedia.org/wiki/Computational_complexity_theory

You can measure how much memory and time an algorithm implementation
needs to run in a specific program on some particular computer, with
specific memory size, other work etc.

When I first got seriously interested in computer performance, about 35
years ago, things were easier. A processor would do one thing, taking
the same amount of time to do it regardless of memory etc., and then go
on to the next thing. The time run a program was simply the sum of the
times for each step. One could usefully estimate the time to run a
program by counting the operations and adding their times.

>
>> Of course, one often needs to know how fast an algorithm will run
>> for a specific problem size on a specific system. In that case,
>> measurement is definitely the way to go.
>
> Well, yes. Now we are on the same page
>
>> With modern computers, any sort of isolated measurement can be
>> misleading. There is so much caching and prediction that code can
>> behave very differently in different contexts.
>
> Yes, Patricia. But give a little thinking to the type of measurement
> I am talking about. Right on top of all system calls and under your
> logical stack. There can and should be utilities to automate this

I will happily think about it once I understand what you mean.

Patricia

Back to comp.lang.java.programmer | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

baseline performance test using java ... lbrt chx _ gemale kom - 2011-07-03 17:53 +0000
  Re: baseline performance test using java ... Eric Sosman <esosman@ieee-dot-org.invalid> - 2011-07-03 14:08 -0400
  Re: baseline performance test using java ... Abu Yahya <abu_yahya@invalid.com> - 2011-07-04 00:03 +0530
    Re: baseline performance test using java ... Patricia Shanahan <pats@acm.org> - 2011-07-03 11:45 -0700
      Re: baseline performance test using java ... Abu Yahya <abu_yahya@invalid.com> - 2011-07-04 06:52 +0530
        Re: baseline performance test using java ... lewbloch <lewbloch@gmail.com> - 2011-07-04 03:53 -0700
    Re: baseline performance test using java ... lewbloch <lewbloch@gmail.com> - 2011-07-04 03:44 -0700
  Re: baseline performance test using java ... Patricia Shanahan <pats@acm.org> - 2011-07-03 12:09 -0700
  Re: baseline performance test using java ... Patricia Shanahan <pats@acm.org> - 2011-07-04 09:54 -0700

csiph-web