def network_optimization_plot(dir: str, graphs: List[Any]) -> None:
df_all = read_iperf(
os.path.join(os.path.realpath(dir), "iperf-all-on-latest.tsv"),
"offloads+\nzerocopy")
df_offload = read_iperf(
os.path.join(os.path.realpath(dir), "iperf-offload_off-latest.tsv"),
"no offloads")
df_zcopy = read_iperf(
os.path.join(os.path.realpath(dir), "iperf-zerocopy_off-latest.tsv"),
"no zerocopy")
df = pd.concat([df_all, df_offload, df_zcopy])
g = catplot(
data=apply_aliases(df),
x=column_alias("type"),
y=column_alias("iperf-throughput"),
kind="bar",
height=2.5,
# aspect=1.2,
color="black",
palette=None,
)
apply_to_graphs(g.ax, False, -1, 0.18)
graphs.append(g)
def mysql_throughput_graph(df: pd.DataFrame) -> Any:
df = df[["system", "SQL statistics transactions", "General statistics total time"]]
df["General statistics total time"] = df["General statistics total time"].apply(
lambda x: float(x.replace("s", ""))
)
df["mysql-throughput"] = (
df["SQL statistics transactions"] / df["General statistics total time"]
)
df = apply_aliases(df)
g = catplot(
data=df,
x=column_alias("system"),
y=column_alias("mysql-throughput"),
kind="bar",
height=2.5,
# aspect=1.2,
color=color,
palette=None,
order=systems_order(df),
)
apply_to_graphs(g.ax, False, -1, 0.285)
return g
def nginx_graph(df: pd.DataFrame, metric: str) -> Any:
plot_col = ["system"]
thru_ylabel = None
width = None
if metric == "lat":
plot_col.append("lat_avg(ms)")
width = 0.25
elif metric == "thru":
plot_col.append("req_sec_tot")
width = 0.3
plot_df = df[plot_col]
groups = len(set((list(plot_df["system"].values))))
plot_df = apply_aliases(plot_df)
g = catplot(data=plot_df,
x=plot_df.columns[0],
y=plot_df.columns[1],
kind="bar",
order=systems_order(plot_df),
height=2.5,
legend=False,
palette=None,
color="black"
# aspect=1.2,
)
apply_to_graphs(g.ax, False, -1, width)
if metric == "thru":
thru_ylabel = g.ax.get_yticklabels()
thru_ylabel = [str(int(int(x.get_text()))) for x in thru_ylabel]
thru_ylabel[0] = "0"
g.ax.set_yticklabels(thru_ylabel)
return g
def syscalls_perf_graph(df: pd.DataFrame) -> Any:
df2 = df[(df.data_size == 32) & (df.threads == 8)]
df2 = df2.assign(
time_per_syscall=10e6 * df2.total_time / (df2.packets_per_thread * df2.threads)
)
g = catplot(
data=apply_aliases(df2),
x=column_alias("system"),
y=column_alias("time_per_syscall"),
order=systems_order(df2),
kind="bar",
height=2.5,
# aspect=1.2,
color=color,
palette=None,
)
# change_width(g.ax, 0.405)
# g.ax.set_xlabel("")
# g.ax.set_ylabel(g.ax.get_ylabel(), fontsize=8)
# g.ax.set_xticklabels(g.ax.get_xmajorticklabels(), fontsize=8)
# g.ax.set_yticklabels(g.ax.get_ymajorticklabels(), fontsize=8)
# g.ax.grid(which="major")
apply_to_graphs(g.ax, False, -1, 0.28)
return g
def sqlite_graph(df: pd.DataFrame) -> Any:
plot_df = df
df = df[df.columns[2]]
df = df.map(apply_sqlite_rows)
plot_df.update(df)
plot_df["sqlite-op-type"] = plot_df["sqlite-op-type"].map({
"5000 INSERTs into table with no index":
"Insert",
"5000 UPDATES of individual rows":
"Update",
"5000 DELETEs of individual rows":
"Delete",
})
groups = len(set((list(plot_df["system"].values))))
plot_df = apply_aliases(plot_df)
g = catplot(
data=plot_df,
x=plot_df.columns[0],
y=plot_df.columns[2],
kind="bar",
height=2.5,
# aspect=0.8,
order=systems_order(plot_df),
hue="Operation",
legend=False,
palette=["darkgrey", "gray", "black"],
)
apply_to_graphs(g.ax, True, 3, 0.285)
g.ax.legend(frameon=False)
return g
def smp_plot(dir: str, graphs: List[Any]) -> None:
df = pd.read_csv(os.path.join(os.path.realpath(dir), "smp-latest.tsv"),
sep="\t")
df = pd.melt(df,
id_vars=['cores', 'job'],
value_vars=['read-bw', 'write-bw'],
var_name="operation",
value_name="disk-throughput")
df = df.groupby(["cores", "operation"]).sum().reset_index()
df["disk-throughput"] /= 1024
g = catplot(
data=apply_aliases(df),
x=column_alias("cores"),
y=column_alias("disk-throughput"),
hue=column_alias("operation"),
kind="bar",
height=2.5,
legend=False,
palette=["grey", "black"],
)
# change_width(g.ax, 0.25)
# # g.ax.set_xlabel('')
# g.ax.set_xticklabels(g.ax.get_xmajorticklabels(), fontsize=6)
# g.ax.set_yticklabels(g.ax.get_ymajorticklabels(), fontsize=6)
# g.ax.legend(loc='best', fontsize='small')
apply_to_graphs(g.ax, True, 2, 0.285)
graphs.append(g)
def fio_read_write_graph(df: pd.DataFrame) -> Any:
df = pd.melt(
df,
id_vars=["system", "job"],
value_vars=["read-bw", "write-bw"],
var_name="operation",
value_name="disk-throughput",
)
df = df.groupby(["system", "operation"]).sum().reset_index()
df["disk-throughput"] /= 1024
g = catplot(
data=apply_aliases(df),
x=column_alias("system"),
y=column_alias("disk-throughput"),
hue=column_alias("operation"),
order=systems_order(df),
kind="bar",
height=2.5,
palette=palette,
legend=False,
# aspect=1.2,
)
# change_width(g.ax, 0.405)
# g.ax.set_xlabel("")
# g.ax.legend(loc="center", bbox_to_anchor=(0.5, 1.05), ncol=2, frameon=False)
apply_to_graphs(g.ax, True, 2, 0.285)
# g.ax.grid(which="major")
return g
def hdparm_graph(df: pd.DataFrame, metric: str) -> Any:
plot_col = ["system"]
if metric == "cached":
plot_col.append("Timing buffer-cache reads")
df["Timing buffer-cache reads"] = preprocess_hdparm(
df["Timing buffer-cache reads"],
)
elif metric == "buffered":
plot_col.append("Timing buffered disk reads")
df["Timing buffered disk reads"] = preprocess_hdparm(
df["Timing buffered disk reads"],
)
plot_df = df[plot_col]
plot_df = apply_aliases(plot_df)
g = catplot(
data=plot_df,
x=plot_df.columns[0],
y=plot_df.columns[1],
kind="bar",
height=2.5,
# aspect=1.2,
color=color,
palette=None,
)
apply_to_graphs(g.ax, False, -1)
return g
def network_bs_plot(dir: str, graphs: List[Any]) -> None:
df = pd.read_csv(os.path.join(os.path.realpath(dir),
"network-test-bs-latest.tsv"),
sep="\t")
df["network-bs-throughput"] = 1024 / df["time"]
# df["batch_size"] = df["batch_size"].apply(lambda x: str(x)+"KiB")
g = catplot(
data=apply_aliases(df),
x=column_alias("batch_size"),
y=column_alias("network-bs-throughput"),
kind="bar",
height=2.5,
legend=False,
color="black",
palette=None,
)
# change_width(g.ax, 0.25)
# # g.ax.set_xlabel('')
# g.ax.set_xticklabels(g.ax.get_xmajorticklabels(), fontsize=6)
# g.ax.set_yticklabels(g.ax.get_ymajorticklabels(), fontsize=6)
apply_to_graphs(g.ax, False, -1)
graphs.append(g)
def hdparm_zerocopy_plot(dir: str, graphs: List[Any]) -> None:
df_all_on = pd.read_csv(os.path.join(os.path.realpath(dir),
"hdparm-all-on-latest.tsv"),
sep="\t")
df_zcopy_off = pd.read_csv(os.path.join(os.path.realpath(dir),
"hdparm-zerocopy-off-latest.tsv"),
sep="\t")
df_all_on = df_all_on.drop(columns=["system"])
df_zcopy_off = df_zcopy_off.drop(columns=["system"])
df_all_on["Timing buffered disk reads"] = preprocess_hdparm(
df_all_on["Timing buffered disk reads"], )
df_zcopy_off["Timing buffered disk reads"] = preprocess_hdparm(
df_zcopy_off["Timing buffered disk reads"], )
df_all_on["Timing buffer-cache reads"] = preprocess_hdparm(
df_all_on["Timing buffer-cache reads"], )
df_zcopy_off["Timing buffer-cache reads"] = preprocess_hdparm(
df_zcopy_off["Timing buffer-cache reads"], )
df_all_on = df_all_on.T.reset_index()
df_zcopy_off = df_zcopy_off.T.reset_index()
df_zcopy_off.columns = ["hdparm_kind", "hdparm-throughput"]
df_all_on.columns = ["hdparm_kind", "hdparm-throughput"]
df_all_on["feature_spdk"] = pd.Series(["spdk-zerocopy"] *
len(df_all_on.index),
index=df_all_on.index)
df_zcopy_off["feature_spdk"] = pd.Series(["spdk-copy"] *
len(df_zcopy_off.index),
index=df_zcopy_off.index)
plot_df = pd.concat([df_all_on, df_zcopy_off], axis=0)
groups = len(set(list(plot_df["feature_spdk"].values)))
g = catplot(
data=apply_aliases(plot_df),
x=column_alias("feature_spdk"),
y=column_alias("hdparm-throughput"),
kind="bar",
height=2.5,
legend=False,
hue=column_alias("hdparm_kind"),
palette=["grey", "black"],
)
# apply_hatch(groups, g, True)
# change_width(g.ax, 0.25)
# g.ax.set_xlabel("")
# g.ax.set_xticklabels(g.ax.get_xmajorticklabels(), fontsize=6)
# g.ax.set_yticklabels(g.ax.get_ymajorticklabels(), fontsize=6)
apply_to_graphs(g.ax, True, 2)
graphs.append(g)
def memcpy_graph(df: pd.DataFrame) -> Any:
g = catplot(
data=apply_aliases(df),
x=column_alias("memcpy-size"),
y=column_alias("memcpy-time"),
hue=column_alias("memcpy-kind"),
kind="bar",
height=2.5,
aspect=1.2,
)
return g
def mysql_latency_graph(df: pd.DataFrame) -> Any:
groups = len(set((list(df["system"].values))))
g = catplot(
data=apply_aliases(df),
x=column_alias("system"),
y=column_alias("Latency (ms) avg"),
kind="bar",
height=2.5,
# aspect=1.2,
color=color,
palette=None,
order=systems_order(df),
)
apply_to_graphs(g.ax, False, -1, 0.285)
return g
def spdk_zerocopy_plot(dir: str, graphs: List[Any]) -> None:
df = pd.read_csv(os.path.join(os.path.realpath(dir),
"spdk-zerocopy-latest.tsv"),
sep="\t")
df = df.assign(aesnithroughput=df.bytes / df.time / 1024 / 1024)
g = catplot(
data=apply_aliases(df),
x=column_alias("type"),
y=column_alias("aesnithroughput"),
kind="bar",
height=2.5,
aspect=1.2,
)
change_width(g.ax, 0.25)
g.ax.set_xlabel('')
graphs.append(g)
def aesni_plot(dir: str, graphs: List[Any]) -> None:
df = pd.read_csv(os.path.join(os.path.realpath(dir), "aesni-latest.tsv"),
sep="\t")
df = df.assign(aesnithroughput=df.bytes / df.time / 1024 / 1024)
g = catplot(data=apply_aliases(df),
x=column_alias("type"),
y=column_alias("aesnithroughput"),
kind="bar",
height=2.5,
aspect=1.2,
color="black",
palette=None)
apply_to_graphs(g.ax, False, -1, 0.1)
g.ax.set_ylabel(g.ax.get_ylabel(), size=8)
g.ax.set_xticklabels(g.ax.get_xticklabels(), size=8)
g.ax.set_yticklabels(g.ax.get_yticklabels(), size=8)
graphs.append(g)
def iperf_graph(df: pd.DataFrame) -> Any:
df = df[df["direction"] == "send"]
df["iperf-throughput"] = df["bytes"] / df["seconds"] * 8 / 1e9
g = catplot(
data=apply_aliases(df),
x=column_alias("system"),
order=systems_order(df),
y=column_alias("iperf-throughput"),
kind="bar",
height=2.5,
# aspect=1.2,
color=color,
palette=None,
)
# change_width(g.ax, 0.405)
# g.ax.set_xlabel("")
apply_to_graphs(g.ax, False, -1, 0.285)
# g.ax.grid(which="major")
return g
def storage_bs_plot(dir: str, graphs: List[Any]) -> None:
df = pd.read_csv(os.path.join(os.path.realpath(dir), "simpleio-unenc.tsv"),
sep="\t")
df["storage-bs-throughput"] = (10 * 1024) / df["time"]
g = catplot(
data=apply_aliases(df),
x=column_alias("batch-size"),
y=column_alias("storage-bs-throughput"),
kind="bar",
height=2.5,
legend=False,
color="black",
palette=None,
)
# change_width(g.ax, 0.25)
# # g.ax.set_xlabel('')
# g.ax.set_xticklabels(g.ax.get_xmajorticklabels(), fontsize=6)
# g.ax.set_yticklabels(g.ax.get_ymajorticklabels(), fontsize=6)
apply_to_graphs(g.ax, False, -1)
graphs.append(g)
def redis_graph(df: pd.DataFrame, metric: str) -> Any:
df_flag = None
hue = None
col_name = None
legend = False
color = None
palette = None
n_cols = None
thru_ylabel = None
width = None
if metric == "thru":
df_flag = df["metric"] == "Throughput(ops/sec)"
col_name = "Throughput(ops/sec)"
width = 0.285
legend = False
elif metric == "lat":
df_flag = (df["metric"]
== "AverageLatency(us)") & (df["operation"] != "[CLEANUP]")
col_name = "AverageLatency(us)"
hue = "operation"
legend = True
width = 0.285
plot_df = df[df_flag]
groups = len(set((list(plot_df["system"].values))))
plot_df = plot_df.drop(["metric"], axis=1)
plot_df = plot_df.rename(columns={"value": col_name})
plot_df = apply_aliases(plot_df)
if hue is None:
color = "black"
palette = None
else:
palette = ["grey", "black"]
n_cols = 2
g = catplot(
data=plot_df,
x=plot_df.columns[0],
y=plot_df.columns[-1],
hue=hue,
kind="bar",
height=2.5,
order=systems_order(plot_df),
# aspect=1.2,
legend=False,
palette=palette,
color=color,
)
apply_to_graphs(g.ax, legend, n_cols, width)
g.ax.legend(ncol=2,
loc="upper center",
bbox_to_anchor=(0.5, 1.05),
frameon=False)
if metric == "thru":
thru_ylabel = g.ax.get_yticklabels()
thru_ylabel = [
str(float(float(x.get_text()) / 1000)) + "k" for x in thru_ylabel
]
thru_ylabel[0] = "0"
thru_ylabel = [x.replace(".0k", "k") for x in thru_ylabel]
g.ax.set_yticklabels(thru_ylabel)
return g
