Benchmarking is something of an open ended problem - the important
thing, I guess, is that some benchmarks are better than no benchmarks.

That said, the test/*.ijs files, in jsource (except *t.ijs) are
probably worth timing. (The *t.ijs are special and are really about
comparing alternate code paths within a build and - because of the
wide variety of hardware out there - would probably make more sense as
a post install mechanism to switch on/off implementation alternatives
to tune it to the local hardware. But that would be a separate
project...)

Thanks,

-- 
Raul


On Thu, May 12, 2016 at 7:05 AM, Henry Rich <[email protected]> wrote:
> Payoff could mean improved speed, features, or code quality.
>
> In these examples, it means speed mostly, sometimes code quality when
> indicated.
>
> The targets for work were chosen by analysis of a task that processed a
> LaTeX file.
>
> I have been comparing results using a benchmark of running lint on the
> source for lint: a parsing-heavy task that takes a couple of seconds.
>
>
> The reference-count work is intended to address an issue we have all seen:
> that arrays with many boxes have considerable overhead even in just
> referring to the array.
>
> The reference-header work is intended to make (,) in sentences like (x { ,
> y) take small constant time.
>
>
> I feel pretty certain that both of those changes have a payoff that is worth
> the effort of implementation.
>
> For anything beyond that, we will need to be guided by analysis of
> benchmarks to make sure we don't make changes that aren't worthwhile.
>
> I would really like to develop a rich benchmark suite to use for
> measurement.  Suggestions welcomed.
>
> Henry Rich
>
>
> On 5/12/2016 5:39 AM, Raul Miller wrote:
>>
>> I am curious how payoff is measured (I see the estimates, I mean - are
>> the payoffs decreased memory use? increased speed? on either of those
>> do we have decent benchmarks? are they increased code stability?
>> opportunities for better documentation? something else?)
>>
>> I have seen many projects adopt "improved efficiency" measures which
>> instead made the system slower and more fragile. I hope we can avoid
>> that mistake here.
>>
>> (I also I hope I am not sticking my nose somewhere where it causes
>> problems.
>>
>> Thanks,
>>
>
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to