Databases
StackOverflow2013
Postsdbo
POPython multiprocessing performance
 

Note that there are some explanatory texts on larger screens.

plurals
  1. POPython multiprocessing performance
    primarykey
    data
    text
    <p>This should be my third and final question regarding my attempts to increase performance on some statistical analysis that I am doing with python. I have 2 versions of my code (single core vs multiprocessing), I was expecting to gain performance by using multiple cores as I expect my code to uncompress/unpack quite a few binary strings , sadly I noticed that the performance actually decreased by using multiple cores.</p> <p>I am wondering if anyone has a possible explanation for what I observe (<strong>scroll down to the April 16th update for more information</strong>)?</p> <p>The key part of program is the function numpy_array (+ decode in multiprocessing), code snippet below (full code accessible via pastebin, further below):</p> <pre><code>def numpy_array(data, peaks): rt_counter=0 for x in peaks: if rt_counter %(len(peaks)/20) == 0: update_progress() peak_counter=0 data_buff=base64.b64decode(x) buff_size=len(data_buff)/4 unpack_format="&gt;%dL" % buff_size index=0 for y in struct.unpack(unpack_format,data_buff): buff1=struct.pack("I",y) buff2=struct.unpack("f",buff1)[0] if (index % 2 == 0): data[rt_counter][1][peak_counter][0]=float(buff2) else: data[rt_counter][1][peak_counter][1]=float(buff2) peak_counter+=1 index+=1 rt_counter+=1 </code></pre> <p>The multiprocessing version performs this with a set of functions, I will display the key 2 below:</p> <pre><code>def tonumpyarray(mp_arr): return np.frombuffer(mp_arr.get_obj()) def numpy_array(shared_arr,peaks): processors=mp.cpu_count() with contextlib.closing(mp.Pool(processes=processors, initializer=pool_init, initargs=(shared_arr, ))) as pool: chunk_size=int(len(peaks)/processors) map_parameters=[] for i in range(processors): counter = i*chunk_size chunk=peaks[i*chunk_size:(i+1)*chunk_size] map_parameters.append((chunk, counter)) pool.map(decode,map_parameters) def decode ((chunk, counter)): data=tonumpyarray(shared_arr).view( [('f0','&lt;f4'), ('f1','&lt;f4',(250000,2))]) for x in chunk: peak_counter=0 data_buff=base64.b64decode(x) buff_size=len(data_buff)/4 unpack_format="&gt;%dL" % buff_size index=0 for y in struct.unpack(unpack_format,data_buff): buff1=struct.pack("I",y) buff2=struct.unpack("f",buff1)[0] #with shared_arr.get_lock(): if (index % 2 == 0): data[counter][1][peak_counter][0]=float(buff2) else: data[counter][1][peak_counter][1]=float(buff2) peak_counter+=1 index+=1 counter+=1 </code></pre> <p>Full program codes can be accessed via these pastebin links</p> <p><a href="http://pastebin.com/tSqc7Ntj" rel="nofollow noreferrer">Pastebin for single core version</a></p> <p><a href="http://pastebin.com/FWD2SMWm" rel="nofollow noreferrer">Pastebin for multiprocessing version</a></p> <p>The performance that I am observing with a file containing 239 timepoints and ~ 180k measurement pairs per timepoint is ~2.5m for single core and ~3.5 for multiprocessing.</p> <p><strong>PS</strong>: The two previous questions (of my first ever attempts at paralellization):</p> <ol> <li><a href="https://stackoverflow.com/q/15966157/1093485">Python multi-processing</a></li> <li><a href="https://stackoverflow.com/q/15976937/1093485">Making my NumPy array shared across processes</a></li> </ol> <p><strong>-- April 16th --</strong></p> <p>I have been profiling my program with the cProfile library (having <code>cProfile.run('main()')</code> in the <code>__main__</code>, which shows that there is 1 step that is slowing everything down:</p> <pre><code>ncalls tottime percall cumtime percall filename:lineno(function) 23 85.859 3.733 85.859 3.733 {method 'acquire' of 'thread.lock' objects} </code></pre> <p>The thing that I do not understand here is that <code>thread.lock</code> objects are used in <code>threading</code> (to my understanding) but should not be used in multiprocessing as each core should run a single thread (besides having it's own locking mechanism), so how is it that this occurs and why does a single call take 3.7 seconds?</p>
    singulars
    1. PO
      singulars
      1. PTAnswer
    1. This table or related slice is empty.
    1. PTQuestion
    1. USCommunity
    1. USBas Jansen
    plurals
    1. PL
      singulars
      1. LTLinked
    2. PL
      singulars
      1. LTLinked
    3. PL
      singulars
      1. LTLinked
    1. PL
      singulars
      1. LTLinked
    1. This table or related slice is empty.
    1. PO
      singulars
      1. PTAnswer
    2. PO
      singulars
      1. PTAnswer
    3. PO
      singulars
      1. PTAnswer
    1. VO
      singulars
      1. POPython multiprocessing performance
      1. US0x90
      1. VTFavorite
    2. VO
      singulars
      1. POPython multiprocessing performance
      1. This table or related slice is empty.
      1. VTUpMod
    3. VO
      singulars
      1. POPython multiprocessing performance
      1. This table or related slice is empty.
      1. VTUpMod
    1. COcan you share the links to your previous questions in this question? and paste the functions you think are important to the question itself
      singulars
      1. POPython multiprocessing performance
      1. US0x90
    2. COOffcourse, let me edit the question.
      singulars
      1. POPython multiprocessing performance
      1. USBas Jansen
    3. COIt very well could have something to do with the GIL. Check out this presentation: http://www.youtube.com/watch?v=Obt-vMVdM8s
      singulars
      1. POPython multiprocessing performance
      1. USalexhb
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload