1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
|
# -*- coding: utf-8 -*-
"""
Created on Tue Sep 13 10:24:09 2016
@author: rmcleod
"""
import numpy as np
import matplotlib.pyplot as plt
import blosc
SHUFFLE = blosc.BITSHUFFLE # Wow BITSHUFFLE IS FAST
nThreadsList = np.arange( 12, 48+1, 6 )
shuffleList = np.arange(0,2+1)
compressorList = np.array( ['lz4', 'zlib', 'zstd' ] )
clevelList = np.arange(1,6+1)
doseLevelList = [ 0.1, 0.25, 0.5, 1.0, 1.5, 2.0, 4.0 ]
testRuns = 1
blockSizeList = np.array( [2**15, 2**16, 2**17, 2**18, 2**19, 2**20, 2**21, 2**22], dtype='int64' )
colorList = ['steelblue', 'forestgreen', 'purple']
optiShape = [len(doseLevelList), len(compressorList), len(clevelList),
len(nThreadsList), len(blockSizeList), testRuns]
cCompress = np.load( "cCompress.npy" )
cBytes = np.load( "cBytes.npy" )
cRatio = np.load( "cRatio.npy" )
cDecompress = np.load( "cDecompress.npy" )
# Average over runs
# (could take std too)
cCompress = np.mean( cCompress, axis=-1 )
cBytes = np.mean( cBytes, axis=-1 )
cRatio = np.mean( cRatio, axis=-1 )
cDecompress = np.mean( cDecompress, axis=-1 )
originalBytes = 50*4096*4096*4 # Guess of size if we saved as float-32
cCompressRate = originalBytes/4 / cCompress / 1E6
cDecompressRate = originalBytes/4 / cDecompress / 1E6
best = np.unravel_index( np.argmax( cCompressRate[3,0,...] ), cCompressRate.shape[2:] )
print( "LZ4 best compression rate: %f at clevel=%d, n_threads=%d, blocksize=%d" % \
(np.max( cCompressRate[3,0,...] ), clevelList[best[0]], nThreadsList[best[1]], blockSizeList[best[2]] ) )
best = np.unravel_index( np.argmax( cCompressRate[3,1,...] ), cCompressRate.shape[2:] )
print( "Zlib best compression rate: %f at clevel=%d, n_threads=%d, blocksize=%d" % \
(np.max( cCompressRate[3,1,...] ), clevelList[best[0]], nThreadsList[best[1]], blockSizeList[best[2]] ) )
best = np.unravel_index( np.argmax( cCompressRate[3,2,...] ), cCompressRate.shape[2:] )
print( "Zstd best compression rate: %f at clevel=%d, n_threads=%d, blocksize=%d" % \
(np.max( cCompressRate[3,2,...] ), clevelList[best[0]], nThreadsList[best[1]], blockSizeList[best[2]] ) )
#for I, doseLevel in enumerate( doseLevelList ):
# for J, compressor in enumerate(compressorList):
# for K, clevel in enumerate( clevelList ):
# for N, N_THREADS in enumerate( nThreadsList ):
# for B, blocksize in enumerate( blockSizeList ):
#### Compression rate ####
maxLz4cRate = np.max( np.max( np.max( cCompressRate[:,0,...], axis=3 ), axis=2 ), axis=1 )
maxZLibcRate = np.max( np.max( np.max( cCompressRate[:,1,...], axis=3 ), axis=2 ), axis=1 )
maxZStdcRate = np.max( np.max( np.max( cCompressRate[:,2,...], axis=3 ), axis=2 ), axis=1 )
maxLz4deRate = np.max( np.max( np.max( cDecompressRate[:,0,...], axis=3 ), axis=2 ), axis=1 )
maxZLibdeRate = np.max( np.max( np.max( cDecompressRate[:,1,...], axis=3 ), axis=2 ), axis=1 )
maxZStddeRate = np.max( np.max( np.max( cDecompressRate[:,2,...], axis=3 ), axis=2 ), axis=1 )
# Something wrong with the decompression benchmark...
uint4CRate = originalBytes/4 / 0.8 / 1E6
plt.figure()
plt.plot( doseLevelList, maxLz4cRate, '.-',
linewidth=1.5, markeredgecolor='k', color=colorList[0], label='lz4' )
plt.plot( doseLevelList, maxLz4deRate, '--', color=colorList[0], label='unlz4' )
plt.plot( doseLevelList, maxZLibcRate, '.-',
linewidth=1.5, markeredgecolor='k', color=colorList[1], label='zlib' )
plt.plot( doseLevelList, maxZLibdeRate, '--', color=colorList[1], label='unzlib' )
plt.plot( doseLevelList, maxZStdcRate, '.-',
linewidth=1.5, markeredgecolor='k', color=colorList[2], label='zstd' )
plt.plot( doseLevelList, maxZStddeRate, '--', color=colorList[2], label='unzstd' )
plt.plot( doseLevelList, uint4CRate*np.ones_like(doseLevelList), '-',
linewidth=1.5, color='darkorange', label='uint4' )
plt.legend( loc='best' )
plt.xlabel( "Dose ($e^-/pix/frame$)" )
plt.ylabel( "Compression Rate ($MB/s$)" )
plt.savefig( "DoseRate_vs_CompressionRate.png" )
#### Compression ratio ####
maxLz4cRatio = np.max( np.max( np.max( cRatio[:,0,...], axis=3 ), axis=2 ), axis=1 )
maxZLibcRatio = np.max( np.max( np.max( cRatio[:,1,...], axis=3 ), axis=2 ), axis=1 )
maxZStdcRatio = np.max( np.max( np.max( cRatio[:,2,...], axis=3 ), axis=2 ), axis=1 )
plt.figure()
plt.plot( doseLevelList, maxLz4cRatio, '.-',
linewidth=1.5, markeredgecolor='k', color=colorList[0], label='lz4' )
plt.plot( doseLevelList, maxZLibcRatio, '.-',
linewidth=1.5, markeredgecolor='k', color=colorList[1], label='zlib' )
plt.plot( doseLevelList, maxZStdcRatio, '.-',
linewidth=1.5, markeredgecolor='k', color=colorList[2], label='zstd' )
plt.plot( doseLevelList, 800.0 * np.ones_like(doseLevelList), '-',
linewidth=1.5, color='darkorange', label='uint4' )
plt.legend( loc='best' )
plt.xlabel( "Dose ($e^-/pix/frame$)" )
plt.ylabel( "Compression Ratio ($\%$)" )
plt.savefig( "DoseRate_vs_CompressionRatio.png" )
# Compression Rate, Ratio, versus clevel
# Just plot for clevel 1?
for I, doseLevel in enumerate( doseLevelList ):
fig = plt.figure()
ax = fig.add_subplot(111)
for J, compressor in enumerate( compressorList ):
bestCompressRate = np.zeros_like( clevelList )
optCRatio = np.zeros_like( clevelList )
for K in np.arange( len( clevelList ) ):
bestNB = np.unravel_index( np.argmax( cCompressRate[I,J,K,...] ) , cCompressRate.shape[-2:] )
bestCompressRate[K] = cCompressRate[I,J,K,bestNB[0],bestNB[1]]
optCRatio[K] = cRatio[I,J,K,bestNB[0], bestNB[1]]
plt.semilogx( bestCompressRate, optCRatio, '.-', color=colorList[J], label=compressorList[J],
linewidth=1.5, markeredgecolor='k' )
ax.annotate( '1', xy=(bestCompressRate[0],optCRatio[0]) )
plt.legend( loc='best' )
plt.xlabel( "Compression Rate ($MB/s$)" )
plt.ylabel( "Compression Ratio ($\%$)" )
plt.title( "Dose level $%.1f e^-/pix/frame$" % doseLevelList[I] )
plt.savefig( "clevelPlot_dose%f.png" % doseLevelList[I] )
#### FIND BEST COMPRESSION LEVEL, NTHREADS, BLOCKSIZE FOR COMPRESSIONRATE AND CRATIO
# I think scaling by the number of threads is a little too confusing?
fig = plt.figure()
I = 3 # 1.0 e-/pix
K = 0
B = 1
for J, compressor in enumerate( compressorList ):
plt.plot( nThreadsList, cCompressRate[I,J,K,:,B], '.-', color=colorList[J], label=compressorList[J],
linewidth=1.5, markeredgecolor='k' )
plt.title( "Thread scaling for $1.0 e^-/pix/frame$, clevel $1$, block size $64k$" )
plt.xlabel( "Number of threads" )
plt.ylabel( "Compression Rate ($MB/s$)" )
plt.legend( loc='best' )
plt.savefig( "CompressionRate_threadScaling.png" )
fig = plt.figure()
I = 3 # 1.0 e-/pix
K = 0
B = 1
for J, compressor in enumerate( compressorList ):
normCompressRate = cCompressRate[I,J,K,:,B] / np.array( nThreadsList )
plt.plot( nThreadsList, normCompressRate, '.-', color=colorList[J], label=compressorList[J],
linewidth=1.5, markeredgecolor='k' )
plt.title( "Thread scaling for $1.0 e^-/pix/frame$, clevel $1$, block size $64k$" )
plt.xlabel( "Number of threads" )
plt.ylabel( "Normalized Compression Rate ($MB/s/thread$)" )
plt.legend( loc='best' )
plt.savefig( "NormedCompressionRate_threadScaling.png" )
fig, ax1 = plt.subplots( figsize=(10,8) )
I = 3 # 1.0 e-/pix
K = 0
N = 5
for J, compressor in enumerate( compressorList ):
ax1.plot( blockSizeList/(2**10), cRatio[I,J,K,N,:], '+-', color=colorList[J], label="Ratio " + compressorList[J],
linewidth=1.5, markeredgecolor='k' )
ax1.set_xlabel( "Block size ($kB$)" )
ax1.set_ylabel( "Compression Ratio ($\%$)" )
ax1.legend( loc='center' )
ax2 = ax1.twinx()
for J, compressor in enumerate( compressorList ):
ax2.plot( blockSizeList/(2**10), cCompressRate[I,J,K,N,:], 'o--', color=colorList[J], label="Rate " + compressorList[J],
linewidth=1.5, markeredgecolor='k' )
ax2.set_ylabel( "Compression Rate ($MB/s$)" )
plt.title( "Scaling with block size" )
ax2.legend( loc='right' )
plt.savefig( "blockScaling.png" )
#plt.figure()
#plt.plot( doseLevelList, compressStackSize, 'k.-' )
#plt.xlabel( r"Dose Rate $(e^-/\AA^2)$" )
#plt.ylabel( r"Compressed file size (GB)" )
#plt.title( r"Expected stack size for a total dose of $80 e^-/\AA^2 (zstd5)$" )
#plt.savefig( "ExpectedMRCZSizeFor80e.png" )
|
