def bench(self, BSIZES=(
128,
250,
2048,
5000,
2**13,
2**20,
)):
from pyutil import benchutil
funcs = (
"_benchmark_insert",
"_benchmark_init_and_has_key_and_del",
"_benchmark_init_and_remove",
"_benchmark_init_and_popitem",
"_benchmark_update",
"_benchmark_init",
)
max = 0
for func in funcs:
if len(func) > max:
max = len(func)
for func in funcs:
print func + " " * (max + 1 - len(func))
for BSIZE in BSIZES:
f = getattr(self, func)
benchutil.rep_bench(f,
BSIZE,
self._generic_benchmarking_init,
MAXREPS=self.MAXREPS,
MAXTIME=self.MAXTIME)
def run_benchmarks(self, profile=False):
for (initfunc, func) in [(self.init_for_unpack, self.unpack),
(self.init_for_pack, self.pack),
(self.init_for_unpack, self.unpack_and_repack)]:
print "benchmarking %s" % (func,)
for N in 16, 512, 2048, 16384:
print "%5d" % N,
benchutil.rep_bench(func, N, initfunc=initfunc, MAXREPS=20, UNITS_PER_SECOND=1000)
benchutil.print_bench_footer(UNITS_PER_SECOND=1000)
print "(milliseconds)"
def bench(self, BSIZES=(128, 250, 2048, 5000, 2**13, 2**20,)):
from pyutil import benchutil
funcs = ("_benchmark_insert", "_benchmark_init_and_has_key_and_del", "_benchmark_init_and_remove", "_benchmark_init_and_popitem", "_benchmark_update", "_benchmark_init",)
max = 0
for func in funcs:
if len(func) > max:
max = len(func)
for func in funcs:
print func + " " * (max + 1 - len(func))
for BSIZE in BSIZES:
f = getattr(self, func)
benchutil.rep_bench(f, BSIZE, self._generic_benchmarking_init, MAXREPS=self.MAXREPS, MAXTIME=self.MAXTIME)
def bench(k, m):
SIZE = 10**6
MAXREPS = 64
# for f in [_encode_file_stringy_easyfec, _encode_file_stringy, _encode_file, _encode_file_not_really,]:
# for f in [_encode_file,]:
# for f in [_encode_file_not_really, _encode_file_not_really_and_hash, _encode_file, _encode_file_and_hash,]:
# for f in [_encode_data_not_really, _encode_data_easyfec, _encode_data_fec,]:
print "measuring encoding of data with K=%d, M=%d, reporting results in nanoseconds per byte after encoding %d bytes %d times in a row..." % (
k, m, SIZE, MAXREPS)
# for f in [_encode_data_fec, _encode_data_not_really]:
for f in [_encode_data_fec]:
def _init_func(size):
return _make_new_rand_data(size, k, m)
for BSIZE in [SIZE]:
results = benchutil.rep_bench(f,
n=BSIZE,
initfunc=_init_func,
MAXREPS=MAXREPS,
MAXTIME=None,
UNITS_PER_SECOND=1000000000)
print "and now represented in MB/s..."
print
best = results['best']
mean = results['mean']
worst = results['worst']
print "best: % 4.3f MB/sec" % (10**3 / best)
print "mean: % 4.3f MB/sec" % (10**3 / mean)
print "worst: % 4.3f MB/sec" % (10**3 / worst)
def byte_at_a_time_cracker(test_guess, secretlen, alphabetsize):
# If we were cleverer, we'd add some backtracking behaviour where, if we can't find any x such that ABCx stands out from the crowd as taking longer than all the other ABCy's, then we start to think that we've taken a wrong step and we go back to trying ABy's. Make sense? But we're not that clever. Once we take a step, we don't backtrack.
print()
guess = []
while len(guess) < secretlen:
best_next_byte = None
best_next_byte_time = None
# For each possible byte...
for next_byte in range(alphabetsize):
c = chr(next_byte)
# Construct a guess with our best candidate so far...
candidate_guess = guess[:]
# Plus that byte...
candidate_guess.append(c)
s = ''.join(candidate_guess)
# Plus random bytes...
s += os.urandom(32 - len(s))
# And see how long it takes the test_guess to consider it...
def f(n):
for i in range(n):
test_guess(s)
times = benchutil.rep_bench(f, 10**7, MAXREPS=10**3, quiet=True)
fastesttime = times['mean']
print("%s..." % (c.encode('hex'), ), end=' ')
if best_next_byte is None or fastesttime > best_next_byte_time:
print("new candidate for slowest next-char: %s, took: %s" % (
c.encode('hex'),
fastesttime,
),
end=' ')
best_next_byte_time = fastesttime
best_next_byte = c
# Okay we've tried all possible next bytes. Our guess is this one (the one that took longest to be tested by test_guess):
guess.append(best_next_byte)
print("SLOWEST next-char %s! Current guess at secret: %s" % (
best_next_byte.encode('hex'),
''.join(guess).encode('hex'),
))
guess = ''.join(guess)
print("Our guess for the secret: %r" % (guess, ))
return guess
def print_measurements():
N=10**4
REPS=10**2
print "all times are in nanoseconds per comparison (in scientific notation)"
print
for comparator in [eqeqcomp, hashcomp, xorcomp, sillycomp]:
print "using comparator ", comparator
# for (a, b, desc) in [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]:
trials = [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]
random.shuffle(trials)
for (a, b, desc) in trials:
print "comparing two strings that are %s to each other" % (desc,)
def f(n):
compare(n, comparator, a, b)
benchutil.rep_bench(f, N, UNITS_PER_SECOND=10**9, MAXREPS=REPS)
print
def print_measurements():
N = 10**4
REPS = 10**2
print "all times are in nanoseconds per comparison (in scientific notation)"
print
for comparator in [eqeqcomp, hashcomp, xorcomp, sillycomp]:
print "using comparator ", comparator
# for (a, b, desc) in [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]:
trials = [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]
random.shuffle(trials)
for (a, b, desc) in trials:
print "comparing two strings that are %s to each other" % (desc, )
def f(n):
compare(n, comparator, a, b)
benchutil.rep_bench(f, N, UNITS_PER_SECOND=10**9, MAXREPS=REPS)
print
def byte_at_a_time_cracker(test_guess, secretlen, alphabetsize):
# If we were cleverer, we'd add some backtracking behaviour where, if we can't find any x such that ABCx stands out from the crowd as taking longer than all the other ABCy's, then we start to think that we've taken a wrong step and we go back to trying ABy's. Make sense? But we're not that clever. Once we take a step, we don't backtrack.
print
guess=[]
while len(guess) < secretlen:
best_next_byte = None
best_next_byte_time = None
# For each possible byte...
for next_byte in range(alphabetsize):
c = chr(next_byte)
# Construct a guess with our best candidate so far...
candidate_guess = guess[:]
# Plus that byte...
candidate_guess.append(c)
s = ''.join(candidate_guess)
# Plus random bytes...
s += os.urandom(32 - len(s))
# And see how long it takes the test_guess to consider it...
def f(n):
for i in xrange(n):
test_guess(s)
times = benchutil.rep_bench(f, 10**7, MAXREPS=10**3, quiet=True)
fastesttime = times['mean']
print "%s..."%(c.encode('hex'),),
if best_next_byte is None or fastesttime > best_next_byte_time:
print "new candidate for slowest next-char: %s, took: %s" % (c.encode('hex'), fastesttime,),
best_next_byte_time = fastesttime
best_next_byte = c
# Okay we've tried all possible next bytes. Our guess is this one (the one that took longest to be tested by test_guess):
guess.append(best_next_byte)
print "SLOWEST next-char %s! Current guess at secret: %s" % (best_next_byte.encode('hex'), ''.join(guess).encode('hex'),)
guess = ''.join(guess)
print "Our guess for the secret: %r" % (guess,)
return guess
def eqeqcomp(a, b):
return a == b
def hashcomp(a, b):
salt = os.urandom(32)
return hashlib.md5(salt + a).digest() == hashlib.md5(salt + b).digest()
N = 10**4
REPS = 10**2
print "all times are in nanoseconds per comparison (scientific notation)"
print
for comparator in [eqeqcomp, hashcomp]:
print "using comparator ", comparator
# for (a, b, desc) in [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]:
trials = [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]
random.shuffle(trials)
for (a, b, desc) in trials:
print "comparing two strings that are %s to each other" % (desc, )
def f(n):
compare(n, comparator, a, b)
benchutil.rep_bench(f, N, UNITS_PER_SECOND=10**9, MAXREPS=REPS)
print
# self.stats['get'] = self.stats.get('get', 0) + 1
elif REMOVE_R.search(inline):
mo = REMOVE_R.search(inline)
start = int(mo.group(1))
length = int(mo.group(2))
self.s.remove(start, length)
# self.stats['remove'] = self.stats.get('remove', 0) + 1
elif POP_R.search(inline):
mo = POP_R.search(inline)
start = int(mo.group(1))
length = int(mo.group(2))
self.s.pop(start, length)
# self.stats['pop'] = self.stats.get('pop', 0) + 1
elif INIT_S in inline:
pass
else:
print("Warning, didn't recognize this line: %r" % (inline, ))
count += 1
inline = self.inf.readline()
# print self.stats
benchutil.print_bench_footer(UNITS_PER_SECOND=1000000)
print("(microseconds)")
for N in [600, 6000, 60000]:
b = B(open(sys.argv[1], 'rU'))
print("%7d" % N, end=' ')
benchutil.rep_bench(b.run, N, b.init, UNITS_PER_SECOND=1000000)
f(a, b)
def eqeqcomp(a, b):
return a == b
def hashcomp(a, b):
salt = os.urandom(32)
return hashlib.md5(salt+ a).digest() == hashlib.md5(salt+b).digest()
N=10**4
REPS=10**2
print "all times are in nanoseconds per comparison (scientific notation)"
print
for comparator in [eqeqcomp, hashcomp]:
print "using comparator ", comparator
# for (a, b, desc) in [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]:
trials = [(p1, p1a, 'same'), (p1, p2, 'close'), (p1, p3, 'far')]
random.shuffle(trials)
for (a, b, desc) in trials:
print "comparing two strings that are %s to each other" % (desc,)
def f(n):
compare(n, comparator, a, b)
benchutil.rep_bench(f, N, UNITS_PER_SECOND=10**9, MAXREPS=REPS)
print
d.addCallback(lambda res: self.banana.transport.getvalue())
def f(o):
self._encoded_huge_string = o
d.addCallback(f)
reactor.runUntilCurrent()
def bench_huge_string_decode(self, N):
""" This is actually a test for acceptable performance, and it needs to
be made more explicit, perhaps by being moved into a separate
benchmarking suite instead of living in this test suite. """
o = self._encoded_huge_string
# results = []
self.banana.prepare()
# d.addCallback(results.append)
CHOMP = 4096
for i in range(0, len(o), CHOMP):
self.banana.dataReceived(o[i:i + CHOMP])
# print results
import sys
from twisted.internet import reactor
from pyutil import benchutil
b = B()
for N in 10**3, 10**4, 10**5, 10**6, 10**7:
print "%8d" % N,
sys.stdout.flush()
benchutil.rep_bench(b.bench_huge_string_decode, N, b.setup_huge_string)
# self.stats['get'] = self.stats.get('get', 0) + 1
elif REMOVE_R.search(inline):
mo = REMOVE_R.search(inline)
start = int(mo.group(1))
length = int(mo.group(2))
self.s.remove(start, length)
# self.stats['remove'] = self.stats.get('remove', 0) + 1
elif POP_R.search(inline):
mo = POP_R.search(inline)
start = int(mo.group(1))
length = int(mo.group(2))
self.s.pop(start, length)
# self.stats['pop'] = self.stats.get('pop', 0) + 1
elif INIT_S in inline:
pass
else:
print "Warning, didn't recognize this line: %r" % (inline,)
count += 1
inline = self.inf.readline()
# print self.stats
benchutil.print_bench_footer(UNITS_PER_SECOND=1000000)
print "(microseconds)"
for N in [600, 6000, 60000]:
b = B(open(sys.argv[1], 'rU'))
print "%7d" % N,
benchutil.rep_bench(b.run, N, b.init, UNITS_PER_SECOND=1000000)
self.banana.transport = TestTransport()
self.banana.connectionMade()
d = self.banana.send("a"*N)
d.addCallback(lambda res: self.banana.transport.getvalue())
def f(o):
self._encoded_huge_string = o
d.addCallback(f)
reactor.runUntilCurrent()
def bench_huge_string_decode(self, N):
""" This is actually a test for acceptable performance, and it needs to
be made more explicit, perhaps by being moved into a separate
benchmarking suite instead of living in this test suite. """
o = self._encoded_huge_string
# results = []
self.banana.prepare()
# d.addCallback(results.append)
CHOMP = 4096
for i in range(0, len(o), CHOMP):
self.banana.dataReceived(o[i:i+CHOMP])
# print results
import sys
from twisted.internet import reactor
from pyutil import benchutil
b = B()
for N in 10**3, 10**4, 10**5, 10**6, 10**7:
print "%8d" % N,
sys.stdout.flush()
benchutil.rep_bench(b.bench_huge_string_decode, N, b.setup_huge_string)
