Skip to content

Commit

Permalink
push linux bench results
Browse files Browse the repository at this point in the history
  • Loading branch information
@victoryang00
victoryang00 committed Feb 29, 2024
1 parent b96c877 commit b20bf61
Show file tree
Hide file tree
Showing 8 changed files with 712 additions and 230 deletions.
212 changes: 156 additions & 56 deletions artifact/bench_comparison.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@
"tc",
"bt",
"cg",
"ep",
"ft",
"lu",
"mg",
Expand All @@ -44,7 +43,6 @@
"nas",
"nas",
"nas",
"nas",
"redis",
"hdastar",
]
Expand All @@ -59,8 +57,7 @@
["-g20", "-vn300"],
["-g20", "-vn300"],
["-g20", "-vn300"],
["-g20", "-n300"],
[],
["-g20", "-n1"],
[],
[],
[],
Expand Down Expand Up @@ -88,14 +85,11 @@
"OMP_NUM_THREADS=4",
"OMP_NUM_THREADS=4",
"OMP_NUM_THREADS=4",
"OMP_NUM_THREADS=4",
"a=b",
"a=b",
]

pool = Pool(processes=1)


def run_mvvm():
results = []
results1 = []
Expand Down Expand Up @@ -124,7 +118,7 @@ def run_hcontainer():
results1.append(
pool.apply_async(
common_util.run_hcontainer,
(aot, "linpack", arg[i], envs[i]),
(aot, folder[i], arg[i], envs[i]),
)
)
# print the results
Expand All @@ -133,9 +127,19 @@ def run_hcontainer():
print(exec, output)
lines = output.split("\n")
for line in lines:
if line.__contains__("user"):
exec_time = float(line.split(" ")[0].replace("user", ""))
results.append((exec, exec_time))
if line.__contains__("elapsed"):
# Split the time string into minutes, seconds, and milliseconds
minutes, seconds = line.split(" ")[2].replace("elapsed", "").split(":")
seconds, milliseconds = seconds.split(".")

# Convert each part to seconds (note that milliseconds are converted and added as a fraction of a second)
total_seconds = (
int(minutes) * 60 + int(seconds) + int(milliseconds) / 1000
)

print(total_seconds)
exec_time = total_seconds
results.append((exec, exec_time))
return results


Expand All @@ -148,7 +152,7 @@ def run_qemu_x86_64():
results1.append(
pool.apply_async(
common_util.run_qemu_x86_64,
(aot, "linpack", arg[i], envs[i]),
(aot, folder[i], arg[i], envs[i]),
)
)
# print the results
Expand All @@ -157,9 +161,19 @@ def run_qemu_x86_64():
print(exec, output)
lines = output.split("\n")
for line in lines:
if line.__contains__("user"):
exec_time = float(line.split(" ")[0].replace("user", ""))
results.append((exec, exec_time))
if line.__contains__("elapsed"):
# Split the time string into minutes, seconds, and milliseconds
minutes, seconds = line.split(" ")[2].replace("elapsed", "").split(":")
seconds, milliseconds = seconds.split(".")

# Convert each part to seconds (note that milliseconds are converted and added as a fraction of a second)
total_seconds = (
int(minutes) * 60 + int(seconds) + int(milliseconds) / 1000
)

print(total_seconds)
exec_time = total_seconds
results.append((exec, exec_time))
return results


Expand All @@ -172,18 +186,27 @@ def run_qemu_aarch64():
results1.append(
pool.apply_async(
common_util.run_qemu_aarch64,
(aot, "linpack", arg[i], envs[i]),
(aot, folder[i], arg[i], envs[i]),
)
)
results1 = [x.get() for x in results1]
for exec, output in results1:
print(exec, output)
lines = output.split("\n")
for line in lines:
if line.__contains__("user"):
exec_time = float(line.split(" ")[0].replace("user", ""))
results.append((exec, exec_time))
if line.__contains__("elapsed"):
# Split the time string into minutes, seconds, and milliseconds
minutes, seconds = line.split(" ")[2].replace("elapsed", "").split(":")
seconds, milliseconds = seconds.split(".")

# Convert each part to seconds (note that milliseconds are converted and added as a fraction of a second)
total_seconds = (
int(minutes) * 60 + int(seconds) + int(milliseconds) / 1000
)

print(total_seconds)
exec_time = total_seconds
results.append((exec, exec_time))
return results


Expand All @@ -196,17 +219,27 @@ def run_native():
results1.append(
pool.apply_async(
common_util.run_native,
(aot, "linpack", arg[i], envs[i]),
(aot, folder[i], arg[i], envs[i]),
)
)
results1 = [x.get() for x in results1]
for exec, output in results1:
print(exec, output)
lines = output.split("\n")
for line in lines:
if line.__contains__("user"):
exec_time = float(line.split(" ")[0].replace("user", ""))
results.append((exec, exec_time))
if line.__contains__("elapsed"):
# Split the time string into minutes, seconds, and milliseconds
minutes, seconds = line.split(" ")[2].replace("elapsed", "").split(":")
seconds, milliseconds = seconds.split(".")

# Convert each part to seconds (note that milliseconds are converted and added as a fraction of a second)
total_seconds = (
int(minutes) * 60 + int(seconds) + int(milliseconds) / 1000
)

print(total_seconds)
exec_time = total_seconds
results.append((exec, exec_time))
return results


Expand All @@ -232,56 +265,123 @@ def write_to_csv(filename):
native_results[idx][1],
]
)

def read_from_csv(filename):
with open(filename, "r") as csvfile:
reader = csv.reader(csvfile)
next(reader)
results = []
for row in reader:
results.append((row[0], float(row[1]), float(row[2]), float(row[3]), float(row[4]), float(row[5])))
return results

def plot(results):
# Number of groups and bar width
n = len(results)
bar_width = 0.2
font = {'size': 18}

plt.rc('font', **font)
workloads = defaultdict(list)
for workload,hcontainer_values, mvvm_values, qemu_x86_64_values,qemu_aarch64_values,native_values in results:
workloads[
workload.replace("OMP_NUM_THREADS=", "")
.replace("-g20", "")
.replace("-n300", "")
.replace(" -f ", "")
.replace("-vn300", "")
.replace("maze-6404.txt", "")
.replace("stories110M.bin", "")
.replace("-z tokenizer.bin -t 0.0", "")
.strip()
].append(( hcontainer_values, mvvm_values, qemu_x86_64_values,qemu_aarch64_values,native_values))

index = np.arange(n)
fig, ax = plt.subplots(figsize=(12, 6))
statistics = {}
for workload, times in workloads.items():
hcontainer_values, mvvm_values, qemu_x86_64_values,qemu_aarch64_values,native_values= zip(*times)
statistics[workload] = {
"hcontainer_median": np.median(hcontainer_values),
"mvvm_median": np.median(mvvm_values),
"qemu_x86_64_median" :np.median(qemu_x86_64_values),
"qemu_aarch64_median" :np.median(qemu_aarch64_values),
"native_median" :np.median(native_values),
"hcontainer_std": np.std(hcontainer_values),
"mvvm_std": np.std(mvvm_values),
"qemu_x86_64_std" :np.std(qemu_x86_64_values),
"qemu_aarch64_std" :np.std(qemu_aarch64_values),
"native_std" :np.std(native_values),
}

# Unpack the results for each environment
hcontainer_values, mvvm_values, native_values, qemu_values = zip(*results)
fig, ax = plt.subplots(figsize=(20, 10))
index = np.arange(len(statistics))
bar_width = 0.7/5

# Plotting
bar1 = ax.bar(index, hcontainer_values, bar_width, label="HContainer")
bar2 = ax.bar(index + bar_width, mvvm_values, bar_width, label="MVVM")
bar3 = ax.bar(index + 2 * bar_width, native_values, bar_width, label="Native")
bar4 = ax.bar(index + 3 * bar_width, qemu_values, bar_width, label="QEMU")
for i, (workload, stats) in enumerate(statistics.items()):
ax.bar(
index[i],
stats["hcontainer_median"],
bar_width,
yerr=stats["hcontainer_std"],
capsize=5,
color="blue",
label="hcontainer" if i == 0 else "",
)
ax.bar(
index[i] + bar_width,
stats["mvvm_median"],
bar_width,
yerr=stats["mvvm_std"],
capsize=5,
color="red",
label="mvvm" if i == 0 else "",
)
ax.bar(
index[i]+ bar_width *2,
stats["qemu_x86_64_median"],
bar_width,
yerr=stats["qemu_x86_64_std"],
capsize=5,
color="brown",
label="qemu_x86_64" if i == 0 else "",
)
ax.bar(
index[i]+ bar_width*3,
stats["qemu_aarch64_median"],
bar_width,
yerr=stats["qemu_aarch64_std"],
capsize=5,
color="purple",
label="qemu_aarch64" if i == 0 else "",
)
ax.bar(
index[i]+ bar_width*4,
stats["native_median"],
bar_width,
yerr=stats["native_std"],
capsize=5,
color="cyan",
label="native" if i == 0 else "",
)
# ax.set_xlabel(workload)
ticklabel = (x for x in list(statistics.keys()))
print(statistics.keys())
ax.set_xticks(index)

ax.set_ylabel("Latency (s)")
ax.set_xticks(index + bar_width * 1.5)
ax.set_xticklabels(cmd)
ax.set_xticklabels(ticklabel,fontsize =10)
ax.set_ylabel("Execution time (s)")
ax.legend()

# Display the plot
# plt.show()
# add text at upper left
ax.legend(loc="upper right")

plt.savefig("performance_comparison.pdf")


if __name__ == "__main__":
mvvm_results = run_mvvm()
native_results = run_native()
qemu_x86_64_results = run_qemu_x86_64()
# print the results
# # print the results
qemu_aarch64_results = run_qemu_aarch64()
hcontainer_results = run_hcontainer()

write_to_csv("comparison.csv")
results = []
for idx in range(len(mvvm_results)):
# Assuming the second element of each tuple is the performance metric,
# and that all lists are aligned and of the same length.
combined_result = (
hcontainer_results[idx][1], # HContainer value
mvvm_results[idx][1], # MVVM value
qemu_x86_64_results[idx][1]
+ qemu_aarch64_results[idx][
1
], # Sum of QEMU x86_64 and QEMU AArch64 values
native_results[idx][1], # Native value
)
results.append(combined_result)

results = read_from_csv("comparison.csv")
plot(results)
Loading

0 comments on commit b20bf61

Please sign in to comment.