def reverse_shared(num:int=1, time:float=0.1, pause:float=0.1, vms=None):
assert vms, "vms is a mandatory argument"
result = defaultdict(list)
def measure(vm, r):
ins, cycles = vm.stat(time)
r[vm.bname] = ins / cycles
for i in range(num):
print("measure %s out of %s" % (i+1, num))
for victim in vms:
"""victim is a VM that we are going to freeze"""
shared, exklusiv = {}, {}
# shared phase
threadulator(measure, [(vm, shared) for vm in vms if vm != victim])
# "stop victim" phase
victim.freeze()
threadulator(measure, [(vm, exklusiv) for vm in vms if vm != victim])
victim.unfreeze()
# calculate results
try:
for bench, pShared, pExcl in dictzip(shared, exklusiv):
key = victim.bname, bench
value = pShared / pExcl
result[key].append(value)
except KeyError:
print("something was wrong, we lost a datapoint")
sleep(pause)
return result
def distribution_with_subsampling(data, skip=22, mode='hist'):
standard = data.standard
withskip = data.withskip
plots, labels = [], []
tuples = []
for test, st_ipcs, raw_skp in dictzip(standard, withskip):
skp_ipc = []
for skp in raw_skp:
instr = mean(skp['instructions'][skip:])
cycles = mean(skp['cycles'][skip:])
if not cycles:
continue
skp_ipc.append(instr/cycles)
tuples.append((test, st_ipcs, skp_ipc))
tuples = sorted(tuples, key=lambda x: x[0])
for test, isolated, shared in tuples:
# if prec >0.1: continue
plots.append(isolated)
plots.append(shared)
labels.append(test)
myboxplot(plots, labels=labels,
legend=["frozen env meas. 50ms", "frozen with skip %sms" % (skip*2)])
def compare(data1, data2, legend):
assert data1, "data1 cannot be empty"
assert data2, "data2 cannot be empty"
def ipc(data): return [t[0]/t[1] for t in data]
data, labels = [], []
for test, p, s in sorted(dictzip(data1,data2)):
print(test, p[:2], s[:2])
labels.append(test)
data.append(p)
data.append(ipc(s))
myboxplot(data, labels=labels, legend=legend, notch=True)
def distribution(data, mode='hist'):
data = pickle.load(open(data, 'rb'))
pure = data.result.pure # data from pure performance isolation
quasi = data.result.quasi # data from quasi isolation
if mode == 'hist':
for i, (test, ref, measurements) in enumerate(dictzip(pure, quasi)):
# print(test, len(ref), len(measurements))
print(ref)
p.subplot(4,2,i+1)
p.title(test)
p.hist(ref, bins=50, label="real isolation", alpha=0.7)
p.hist(measurements, bins=50, label="quasi-isolation", alpha=0.7)
p.legend(loc="best")
p.xlim(0,3)
elif mode == 'boxplot':
labels = []
data = []
for i, (test, ref, measurements) in enumerate(sorted(dictzip(pure, quasi))):
labels.append(test)
data.append(ref)
data.append(measurements)
box = myboxplot(data, labels=labels, notch=True)
def analyze_reverse2(data):
data = pickle.load(open(data, 'rb')).result
isolated = data.isolated
shared = data.shared
plots, labels = [], []
tuples = []
for test, isolated, shared in dictzip(isolated, shared):
prec = precision2(shared, isolated)
tuples.append((test, isolated, shared, prec))
tuples = sorted(tuples, key=lambda x: x[3])
for test, isolated, shared, prec in tuples:
# if prec >0.1: continue
plots.append(isolated)
plots.append(shared)
labels.append("{} {:.1%}".format(test, prec))
myboxplot(plots, labels=labels)
def loosers(num:int=10, interval:int=100, pause:float=0.0, vms=None):
""" Detect starving applications. """
shared_perf = defaultdict(list)
frozen_perf = defaultdict(list)
while num>0:
print(num, "measurements left")
shared_sampling(num=10, interval=interval, pause=pause, result=shared_perf, vms=vms)
freezing_sampling(num=10, interval=interval, pause=pause, result=frozen_perf, vms=vms)
num -= 10
result = {}
for bench, sh_perf, fr_perf in dictzip(shared_perf, frozen_perf):
ratio = mean(sh_perf) / mean(fr_perf)
result[bench] = ratio
print("our technique:")
print(sorted(result.items(), key=lambda v: v[1]))
return result
def real_loosers3(num:int=10, interval:int=10*1000, pause:float=0.1, vms=None):
""" Detect starving applications. Improved version """
shared_perf = defaultdict(list)
frozen_perf = defaultdict(list)
prints("real_loosers3 iterations ", end="")
for i,x in enumerate(range(num)):
prints("{i}/{num}", end=" "); sys.stdout.flush()
shared_thr_sampling(num=1, interval=interval, result=shared_perf, vms=vms)
freezing_sampling(num=1, interval=interval, pause=0.1, result=frozen_perf, vms=vms)
sleep(pause)
result = {}
for vm, sh_perf, fr_perf in dictzip(shared_perf, frozen_perf):
ratio = mean(sh_perf) / mean(fr_perf)
result[vm] = ratio
result = sorted(result.items(), key=lambda v: v[1])
print("Loosers: straight forward technique:\n", result)
return result, shared_perf, frozen_perf
