[9fans] some measurements in plan 9

[9fans] some measurements in plan 9

[hugo rivera]

Hi,
I want to compare the memory consumption of two versions of the same
program. I think /proc it's the way to go and acid should give me the
tools to do so, am I right? is there a better way to do so? Just
asking before reading the acid papers.
Also, I am interested in the speed of both versions, and I am planning
to use the time command with some rc scripts. Am I in the right path?
thanks and saludos

--
Hugo

Re: [9fans] some measurements in plan 9

[ron minnich]

On Wed, Apr 15, 2009 at 7:53 AM, hugo rivera <uair00@gmail.com> wrote:
> Hi,
> I want to compare the memory consumption of two versions of the same
> program. I think /proc it's the way to go and acid should give me the
> tools to do so, am I right? is there a better way to do so? Just
> asking before reading the acid papers.

seems reasonable to me, I assume you are looking at data consumption only?

> Also, I am interested in the speed of both versions, and I am planning
> to use the time command with some rc scripts. Am I in the right path?

If they run for hours, time is fine. If they run for seconds, well,
maybe not so fine.

ron

Re: [9fans] some measurements in plan 9

[hugo rivera]

> seems reasonable to me, I assume you are looking at data consumption only?

well, I am not really sure what you mean. Data consumption? ;-)

> If they run for hours, time is fine. If they run for seconds, well,
> maybe not so fine.

OK, they just run for a few seconds, so, any suggestions are welcome
(in fact needed)

--
Hugo

Re: [9fans] some measurements in plan 9

[ron minnich]

On Wed, Apr 15, 2009 at 8:18 AM, hugo rivera <uair00@gmail.com> wrote:
>> seems reasonable to me, I assume you are looking at data consumption only?
>
> well, I am not really sure what you mean. Data consumption? ;-)

sorry. Memory data footprint. Not code + memory. This all depends on
lots of factors, but for code remember that text is shared.


> OK, they just run for a few seconds, so, any suggestions are welcome
> (in fact needed)

/dev/bintime

ron

Re: [9fans] some measurements in plan 9

[Eris Discordia]

>> OK, they just run for a few seconds, so, any suggestions are welcome
>> (in fact needed)
>
> /dev/bintime

Wouldn't many rounds of running with different data sets (to minimize cache
hits) and timing with time also serve the same purpose? Does time introduce
some unavoidable minimum margin of error for short runs that will get
multiplied by the number of times and drown the actual measurement? If time
only introduces fluctuations they will cancel each other out in many runs
but if there's an unavoidable minimum error then multiple runs won't help.

--On Wednesday, April 15, 2009 8:22 AM -0700 ron minnich
<rminnich@gmail.com> wrote:

> On Wed, Apr 15, 2009 at 8:18 AM, hugo rivera <uair00@gmail.com> wrote:
>>> seems reasonable to me, I assume you are looking at data consumption
>>> only?
>>
>> well, I am not really sure what you mean. Data consumption? ;-)
>
> sorry. Memory data footprint. Not code + memory. This all depends on
> lots of factors, but for code remember that text is shared.
>
>
>> OK, they just run for a few seconds, so, any suggestions are welcome
>> (in fact needed)
>
> /dev/bintime
>
> ron
>

Re: [9fans] some measurements in plan 9

[hugo rivera]

many thanks.

2009/4/15 ron minnich <rminnich@gmail.com>:
> On Wed, Apr 15, 2009 at 8:18 AM, hugo rivera <uair00@gmail.com> wrote:
>>> seems reasonable to me, I assume you are looking at data consumption only?
>>
>> well, I am not really sure what you mean. Data consumption? ;-)
>
> sorry. Memory data footprint. Not code + memory. This all depends on
> lots of factors, but for code remember that text is shared.
>
>
>> OK, they just run for a few seconds, so, any suggestions are welcome
>> (in fact needed)
>
> /dev/bintime
>
> ron
>
>



--
Hugo

Re: [9fans] some measurements in plan 9

[ron minnich]

actually i never use anything less than the processor cycle counter if
I care. If I don't care as much bintime is good. time on plan 9 is
great but for runs in the seconds I don't use it.

ron

Re: [9fans] some measurements in plan 9

[erik quanstrom]

On Wed Apr 15 10:55:34 EDT 2009, uair00@gmail.com wrote:
> Hi,
> I want to compare the memory consumption of two versions of the same
> program. I think /proc it's the way to go and acid should give me the
> tools to do so, am I right? is there a better way to do so? Just
> asking before reading the acid papers.
> Also, I am interested in the speed of both versions, and I am planning
> to use the time command with some rc scripts. Am I in the right path?
> thanks and saludos

leak should give you a very detailed picture of memory allocation.

ron's comments notwithstanding, i have found nsec() to be very helpful
in tuning most user mode programs.  i wasn't doing supercomputer
applications, but the chances are you aren't either.

- erik

Re: [9fans] some measurements in plan 9

[hugo rivera]

> leak should give you a very detailed picture of memory allocation.
>
> ron's comments notwithstanding, i have found nsec() to be very helpful
> in tuning most user mode programs.  i wasn't doing supercomputer
> applications, but the chances are you aren't either.

No, no supercomputers for me (maybe one day ;-)

-- 
Hugo