Re: Problem with (Saturn) Profiler
Re: Problem with (Saturn) Profiler
- Subject: Re: Problem with (Saturn) Profiler
- From: Jeffrey Oleander <email@hidden>
- Date: Fri, 30 Jan 2009 08:13:38 -0800 (PST)
> On Fri, 2009/01/30, Joar Wingfors <email@hidden> wrote:
> From: Joar Wingfors <email@hidden>
> Subject: Re: Problem with (Saturn) Profiler
> To: email@hidden
> Cc: "Rick Altherr" <email@hidden>, email@hidden
> Date: Friday, 2009 January 30, 09:41
>> On 2009 Jan 30, at 06:35, Jeffrey Oleander wrote:
>> Or at least that's been my experience. The only
>> down-side is the massive performance over-head
>> while you're running the test and the massive
>> load of data from the vast majority of routines
>> that, once you've run it, you know don't need
>> your attention.
> Wouldn't such "massive overhead" likely affect
> the behavior of your application (at least if you
> have any asynchronicity / concurrency),
Yes, it would, on the backwards "modern" systems,
which don't even have a system clock that will
give remotely accurate millisecond timings, while
the core, itself, runs at nominal giga-Herz
frequencies. Not to mention the overhead of the
asynchronicity mechanisms, themselves.
> making the value of the output questionable, or
> at least difficult to interpret?
> One of the really nice things about Sharks CPU
> samplers is the very low overhead - and therefore
> impact - while gathering performance data.
Nope, there's still an impact due to the loosey
goosey asynchronicity itself, because it would be
very hit or miss what it would catch in its samples,
and therefore very hit or miss whether it would
be deceptive or valuable. Maybe if you monitored
it while a few dozen beta testers hammered away,
you might be able to digest down the data from
their more random behavior to get some believable
numbers. Merely repeating the test using driving
and driven systems, or repeating the test a few
times, wouldn't suffice.
But, still, it's better than nothing. Shark
is not useless.
I used to test performance on super-computers
using the driving/driven system set-up. But,
as someone else pointed out, most of the tuning
and clever (bizarre, mostly) tweaks turn out
to be far more costly for the gain than inspired
redesigns using a different algorithm for what
you're doing. That's where the big performance
gains are.
Shalom Shabbat.
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Xcode-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden