Last active
April 28, 2023 21:43
-
-
Save braingram/ba388b166bc08b2c42ce1abee317f1b0 to your computer and use it in GitHub Desktop.
Performance of memmapping vs chunking (zarr)
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import shutil | |
import time | |
import numpy | |
import zarr | |
n_times = 10 | |
image_size = (4000, 4000, 50) | |
dtype = 'uint8' | |
mm_fn = 'test_memmap.bin' | |
zarr_dir = 'test_zarr' | |
def setup_memmap(): | |
# generate the memmap file | |
arr = numpy.zeros(image_size, dtype) | |
with open(mm_fn, 'wb') as f: | |
f.write(arr.tobytes()) | |
return numpy.memmap(mm_fn, shape=image_size, dtype='uint8') | |
def re_memmap(): | |
return numpy.memmap(mm_fn, shape=image_size, dtype='uint8') | |
def setup_zarr(chunk_shape): | |
# generate the zarr directory store | |
if os.path.exists(zarr_dir): | |
shutil.rmtree(zarr_dir) | |
z = zarr.open(zarr_dir, mode='w', shape=image_size, chunks=chunk_shape, dtype=dtype) | |
z[:] = 0 | |
return z | |
chunk_shapes = [ | |
image_size, # one chunk for whole image | |
[1000, 1000, 50], # first 2 dimensions | |
[4000, 4000, 10], # only z | |
[1000, 1000, 10], # all 3 | |
[300, 300, 1], # very small chunks | |
] | |
accessors = [ | |
((lambda arr: arr[:]), "whole array"), | |
((lambda arr: arr[:, 0, 0]), "first dimension slice"), | |
((lambda arr: arr[0, :, 0]), "second dimension slice"), | |
((lambda arr: arr[0, 0, :]), "third dimension slice"), | |
((lambda arr: arr[10:40, 20:50, 0]), "cutout for 1 z"), | |
((lambda arr: arr[:30, :30, 0]), "cutout aligned to one chunk"), | |
((lambda arr: arr[10:40, 20:50, :]), "chutout for all z"), | |
((lambda arr: arr[100:400, 200:500, 0]), "larger cutout for 1 z"), | |
((lambda arr: arr[:300, :300, 0]), "larger cutout aligned to one chunk"), | |
((lambda arr: arr[100:400, 200:500, :]), "larger cutout for all z"), | |
] | |
def time_it(arr, accessor, n_times, mean=True): | |
ts = [] | |
v = 0 | |
for _ in range(n_times): | |
t0 = time.monotonic() | |
v += numpy.sum(accessor(arr)) | |
t1 = time.monotonic() | |
ts.append(t1 - t0) | |
if not mean: | |
return ts | |
return numpy.mean(ts) | |
if __name__ == '__main__': | |
print(f"Running each test {n_times} times") | |
print(f"Testing image shape: {image_size}") | |
mm = setup_memmap() | |
for cs in chunk_shapes: | |
print(f"\tchunk shape: {cs}") | |
z = setup_zarr(cs) | |
for (a, an) in accessors: | |
# to make this 'fair' make a new memmap every time | |
mm = re_memmap() | |
zt = time_it(z, a, n_times) | |
mmt = time_it(mm, a, n_times) | |
winner = 'memmap' if mmt < zt else 'chunking' | |
print(f"\t\t{winner} win! {an}: chunking={zt}, memmap={mmt}") |
Results for a run with n_times = 1
Running each test 1 times
Testing image shape: (4000, 4000, 50)
chunk shape: (4000, 4000, 50)
memmap win! whole array: chunking=0.21326945899636485, memmap=0.202122041999246
memmap win! first dimension slice: chunking=0.05284383399703074, memmap=0.003369375001057051
memmap win! second dimension slice: chunking=0.051087416999507695, memmap=4.366700886748731e-05
memmap win! third dimension slice: chunking=0.05022858399024699, memmap=3.1291987397708e-05
memmap win! cutout for 1 z: chunking=0.05552812499809079, memmap=6.087501242291182e-05
memmap win! cutout aligned to one chunk: chunking=0.05585804200381972, memmap=6.279099034145474e-05
memmap win! chutout for all z: chunking=0.05316399999719579, memmap=7.2750000981614e-05
memmap win! larger cutout for 1 z: chunking=0.05350100000214297, memmap=0.0007666250021429732
memmap win! larger cutout aligned to one chunk: chunking=0.05246841699408833, memmap=0.0006311250035651028
memmap win! larger cutout for all z: chunking=0.05274008400738239, memmap=0.001432582997949794
chunk shape: [1000, 1000, 50]
memmap win! whole array: chunking=0.2975324580038432, memmap=0.1921742090053158
memmap win! first dimension slice: chunking=0.016103000001749024, memmap=0.0031840830051805824
memmap win! second dimension slice: chunking=0.017106999992392957, memmap=4.279200220480561e-05
memmap win! third dimension slice: chunking=0.0031345830066129565, memmap=2.8416005079634488e-05
memmap win! cutout for 1 z: chunking=0.0030621659971075132, memmap=6.0999998822808266e-05
memmap win! cutout aligned to one chunk: chunking=0.0033969170035561547, memmap=6.09580019954592e-05
memmap win! chutout for all z: chunking=0.006794417000492103, memmap=7.933301094453782e-05
memmap win! larger cutout for 1 z: chunking=0.00476850000268314, memmap=0.0008481660042889416
memmap win! larger cutout aligned to one chunk: chunking=0.005172499993932433, memmap=0.0008740420016692951
memmap win! larger cutout for all z: chunking=0.006961416002013721, memmap=0.0020638749992940575
chunk shape: [4000, 4000, 10]
memmap win! whole array: chunking=0.5640181249909801, memmap=0.1960677500028396
memmap win! first dimension slice: chunking=0.009735750005347654, memmap=0.0032849169947439805
memmap win! second dimension slice: chunking=0.015671082990593277, memmap=4.616699879989028e-05
memmap win! third dimension slice: chunking=0.05057274999853689, memmap=3.2625001040287316e-05
memmap win! cutout for 1 z: chunking=0.010244542005239055, memmap=6.0000005760230124e-05
memmap win! cutout aligned to one chunk: chunking=0.014627208991441876, memmap=5.395899643190205e-05
memmap win! chutout for all z: chunking=0.04832499999611173, memmap=8.091700146906078e-05
memmap win! larger cutout for 1 z: chunking=0.009592207992682233, memmap=0.0006277909997152165
memmap win! larger cutout aligned to one chunk: chunking=0.009771666998858564, memmap=0.0006600000051548705
memmap win! larger cutout for all z: chunking=0.05056187500304077, memmap=0.001403082991600968
chunk shape: [1000, 1000, 10]
memmap win! whole array: chunking=0.5623051669972483, memmap=0.19121000000450294
chunking win! first dimension slice: chunking=0.002491374994860962, memmap=0.002904458000557497
memmap win! second dimension slice: chunking=0.0023209999926621094, memmap=5.154100654181093e-05
memmap win! third dimension slice: chunking=0.0031024999916553497, memmap=2.333299198653549e-05
memmap win! cutout for 1 z: chunking=0.0007400829927064478, memmap=7.183299749158323e-05
memmap win! cutout aligned to one chunk: chunking=0.0008918749954318628, memmap=6.516701250802726e-05
memmap win! chutout for all z: chunking=0.0040137920004781336, memmap=9.458299609832466e-05
memmap win! larger cutout for 1 z: chunking=0.001043583994032815, memmap=0.0010238329996354878
memmap win! larger cutout aligned to one chunk: chunking=0.0011161250004079193, memmap=0.0009834999946178868
memmap win! larger cutout for all z: chunking=0.010943916990072466, memmap=0.002653999996255152
chunk shape: [300, 300, 1]
memmap win! whole array: chunking=1.595065750007052, memmap=0.19998262499575503
chunking win! first dimension slice: chunking=0.001296250004088506, memmap=0.0033537500130478293
memmap win! second dimension slice: chunking=0.0013369590014917776, memmap=4.5958004193380475e-05
memmap win! third dimension slice: chunking=0.004903499997453764, memmap=3.095899592153728e-05
memmap win! cutout for 1 z: chunking=0.0002564169990364462, memmap=6.795799708925188e-05
memmap win! cutout aligned to one chunk: chunking=0.0002527920005377382, memmap=6.558300810866058e-05
memmap win! chutout for all z: chunking=0.005408583005191758, memmap=9.845799650065601e-05
chunking win! larger cutout for 1 z: chunking=0.0006278750079218298, memmap=0.0008047919982345775
chunking win! larger cutout aligned to one chunk: chunking=0.0002954170049633831, memmap=0.0007959579961607233
memmap win! larger cutout for all z: chunking=0.029081792003125884, memmap=0.001844792001065798
Is the mmap accessor actually loading any data into memory?
Is the mmap accessor actually loading any data into memory?
Good point! I updated it to sum the array (and updated the results above).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Results for a run with
n_times = 10