In [36]:
import pandas as pd
from matplotlib import pyplot as plt
%matplotlib notebook
In [2]:
def parse_and_plot(filename):
df = pd.read_csv(filename,header=None)
df /= float(1e6) # convert to milliseconds
ax = df.plot(kind='hist', bins=50, figsize=(15,8))
ax.set_title("Initial Diff Latency")
ax.set_ylabel("Latency (ms)")
ax = df.plot(kind='line',style='-',figsize=(15,8))
ax.set_title("Initial Diff Latency")
ax.set_ylabel("Latency (ms)")
ax.set_xlabel("Sample #")
return df
In [ ]:
FILENAME="data/10_pub_sub_pairs.csv"
df = parse_and_plot(FILENAME)
df.describe()
100 pairs of pub/sub that share a query. Publishers at a rate of 10 msg/sec, runtime of 20 minutes. Each query has a unique key and value
In [3]:
filename="data/forwarding_latency_100pub_to_1sub.csv"
df = pd.read_csv(filename)
df /= float(1e6)
df.plot(kind='line', figsize=(15,8))
Out[3]:
In [15]:
filename="data/1_client_10_pub.csv"
df1 = pd.read_csv(filename, header=None)
df1 /= float(1e6)
df1.plot(style='-',figsize=(15,8))
print df1.quantile(q=.99)
df1.describe()
Out[15]:
We wil run forwrdingLatencyNto1 for clients = 1, 10, 100. In the N > 1 case, we average the N clients together to get a single timeseries for that case, then we will plot them all together
In [17]:
filename="data/10_client_100_pub.csv"
df2 = pd.read_csv(filename, header=None)
df2 /= float(1e6)
df2[0] = df2.mean(axis=1)
print df2[0].quantile(q=.99)
df2[0].plot(style='-',figsize=(15,8))
#df2.describe()
Out[17]:
In [18]:
df1[1] = df2[0]
In [20]:
filename="data/100_client_1000_pub.csv"
df3 = pd.read_csv(filename, header=None)
df3 /= float(1e6)
df3[0] = df3.mean(axis=1)
df3.plot(style='-',figsize=(15,8), legend=False)
print df3[0].quantile(q=.99)
df1[2] = df3[0]
In [39]:
snp = df1[df1 > 3.5].copy()
snp.columns = ['1:10 sub:pub','10:100 sub:pub', '100:1000 sub:pub']
axs = snp.plot(figsize=(15,10), subplots=True)
for ax in axs:
ax.set_ylabel("Latency (ms)")
ax.set_xlabel("Sample (5msg/sec)")
ax.set_yscale('log')
In [14]:
df1
Out[14]:
In [ ]: