[Yt-dev] Parameter file store

[david collins]

Dude.

Looking at time per parameter file read, reading the first 600 steps
from one sim.

Before the fix, it monotonically increased from 0.01 second/read to 3
seconds/read.

After the fix, it's roughly constant at 0.02.  Huge improvement in
performance.  That's really going to speed things up when I make my
core tracking analysis.  Thanks!

d.

On Thu, Mar 5, 2009 at 10:49 PM, Matthew Turk <matthewturk at gmail.com> wrote:
> Any luck?
>
> On Tue, Mar 3, 2009 at 3:19 PM, Matthew Turk <matthewturk at gmail.com> wrote:
>> Hi Dave,
>>
>> Here's a patch:
>>
>> http://paste.enzotools.org/show/68/
>>
>> You can get this with yt_lodgeit.py --download=68 > patch and apply it
>> with patch -p1 < patch .  Conversely, you can download it from the
>> bitbucket repo or the mirror on hg.enzotools.org.  It defaults to
>> keeping the last 300 touched files around.
>>
>> If this works I'll merge back to trunk.  You can test it just by
>> instantiated a crapton of them in a row.
>>
>> -Matt
>>
>> On Tue, Mar 3, 2009 at 2:25 PM, Matthew Turk <matthewturk at gmail.com> wrote:
>>> That's right, it's completely re-creatable.
>>>
>>> On Tue, Mar 3, 2009 at 2:19 PM, david collins <antpuncher at gmail.com> wrote:
>>>> Sure.  So one would need to re-instantiate if something gets bumped off?
>>>>
>>>> d.
>>>>
>>>> On Tue, Mar 3, 2009 at 2:15 PM, Matthew Turk <matthewturk at gmail.com> wrote:
>>>>> Okay, how about this.
>>>>>
>>>>> We make a 'freshness' column in the CSV.  This is a timestamp of the
>>>>> last time the parameter file was used.  This can be the sort field,
>>>>> too.  We'll set a variable -- default it to, say, 300 -- that includes
>>>>> the freshness cutoff in number of parameter files.  Thus, sorted by
>>>>> freshness, we take the last N, where N is set in the parameter file.
>>>>>
>>>>> Sound good?
>>>>>
>>>>> On Tue, Mar 3, 2009 at 1:47 PM, david collins <antpuncher at gmail.com> wrote:
>>>>>>> Wow, you've been analyzing 10,000 parameter files?  Okay.
>>>>>>
>>>>>> Ci.  Making time sequences.  Hundreds/thousands of snapshots from a
>>>>>> dozen sims, and comparing.   (This is why uber)
>>>>>>
>>>>>> Deleting parameter_files.csv is a good workaround for me now, so don't
>>>>>> spend more than 45 seconds on it.  I did this recently, so it's not
>>>>>> huge, only 68 Mb.
>>>>>>
>>>>>> kraken /nics/b/home/collins/.yt/parameter_files.csv
>>>>>>
>>>>>> d.
>>>>>>
>>>>>>>
>>>>>>> One thing I can do is implement a maximum number in the store.  Then
>>>>>>> we can only guarantee objects will be automatically retrieved (without
>>>>>>> having *recently* instantiated the PF) if they were in the last N,
>>>>>>> where N is the number of files we keep in the store.
>>>>>>>
>>>>>>> Can you send me one of your parameter file stores?  Or put it on kraken?
>>>>>>>
>>>>>>> -Matt
>>>>>>> - Show quoted text -
>>>>>>> On Tue, Mar 3, 2009 at 1:04 PM, david collins <antpuncher at gmail.com> wrote:
>>>>>>>> Hey--
>>>>>>>>
>>>>>>>> Not sure if this belongs on the dev list--
>>>>>>>>
>>>>>>>> I've been reading and analyzing large-ish ( O(10,000) ) parameter
>>>>>>>> files, and the parameter_file.csv gets large (300 Mb), which isn't a
>>>>>>>> problem, but when that happens, doing anything (even importing YT)
>>>>>>>> takes forever (> 1 minute to import yt.lagos).
>>>>>>>>
>>>>>>>> I did install the "unboudned growth" fix you sent, still happens.
>>>>>>>>
>>>>>>>> I just noticed that it  happens after your fix yesterday, or I'da said
>>>>>>>> something earlier.
>>>>>>>>
>>>>>>>> d.
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Yt-dev mailing list
>>>>>>> Yt-dev at lists.spacepope.org
>>>>>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>>>>>>>
>>>>>> _______________________________________________
>>>>>> Yt-dev mailing list
>>>>>> Yt-dev at lists.spacepope.org
>>>>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>>>>>>
>>>>> _______________________________________________
>>>>> Yt-dev mailing list
>>>>> Yt-dev at lists.spacepope.org
>>>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>>>>>
>>>> _______________________________________________
>>>> Yt-dev mailing list
>>>> Yt-dev at lists.spacepope.org
>>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>>>>
>>>
>>
> _______________________________________________
> Yt-dev mailing list
> Yt-dev at lists.spacepope.org
> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>