def collect_rebuffer(client_buffer_results, postgres_cursor):
# process InfluxDB data
x = {}
for pt in client_buffer_results['client_buffer']:
expt_id = int(pt['expt_id'])
expt_config = retrieve_expt_config(expt_id, expt_id_cache,
postgres_cursor)
# index x by (abr, cc)
abr_cc = (expt_config['abr'], expt_config['cc'])
if abr_cc not in x:
x[abr_cc] = {}
session = (pt['user'], int(pt['init_id']), pt['channel'],
int(pt['expt_id']))
if session not in x[abr_cc]:
x[abr_cc][session] = {}
x[abr_cc][session]['min_time'] = None
x[abr_cc][session]['max_time'] = None
x[abr_cc][session]['min_cum_rebuf'] = None
x[abr_cc][session]['max_cum_rebuf'] = None
# shorthand variable
y = x[abr_cc][session]
ts = try_parsing_time(pt['time'])
cum_rebuf = float(pt['cum_rebuf'])
if y['min_time'] is None or ts < y['min_time']:
y['min_time'] = ts
if y['max_time'] is None or ts > y['max_time']:
y['max_time'] = ts
if y['min_cum_rebuf'] is None or cum_rebuf < y['min_cum_rebuf']:
y['min_cum_rebuf'] = cum_rebuf
if y['max_cum_rebuf'] is None or cum_rebuf > y['max_cum_rebuf']:
y['max_cum_rebuf'] = cum_rebuf
# calculate rebuffer rate
rebuffer = {}
for abr_cc in x:
total_play = 0
total_rebuf = 0
for session in x[abr_cc]:
# shorthand variable
y = x[abr_cc][session]
total_play += (y['max_time'] - y['min_time']).total_seconds()
total_rebuf += y['max_cum_rebuf'] - y['min_cum_rebuf']
if total_play == 0:
sys.exit('Error: total play time is 0')
rebuf_rate = total_rebuf / total_play
rebuffer[abr_cc] = rebuf_rate * 100
return rebuffer
def calculate_trans_times(video_sent_results, video_acked_results, cc,
postgres_cursor):
d = {}
last_video_ts = {}
for pt in video_sent_results['video_sent']:
expt_id = int(pt['expt_id'])
session = (pt['user'], int(pt['init_id']), pt['channel'], expt_id)
# filter data points by congestion control
expt_config = retrieve_expt_config(expt_id, expt_id_cache,
postgres_cursor)
if cc is not None and expt_config['cc'] != cc:
continue
if session not in d:
d[session] = {}
last_video_ts[session] = None
video_ts = int(pt['video_ts'])
if last_video_ts[session] is not None:
if video_ts != last_video_ts[session] + VIDEO_DURATION:
sys.stderr.write('Warning in session {}: video_ts={}\n'.format(
session, video_ts))
continue
last_video_ts[session] = video_ts
d[session][video_ts] = {}
dsv = d[session][video_ts] # short name
dsv['sent_ts'] = try_parsing_time(pt['time'])
dsv['size'] = float(pt['size']) / PKT_BYTES # bytes -> packets
# byte/second -> packet/second
dsv['delivery_rate'] = float(pt['delivery_rate']) / PKT_BYTES
dsv['cwnd'] = float(pt['cwnd'])
dsv['in_flight'] = float(pt['in_flight'])
dsv['min_rtt'] = float(pt['min_rtt']) / MILLION # us -> s
dsv['rtt'] = float(pt['rtt']) / MILLION # us -> s
# dsv['ssim_index'] = get_ssim_index(pt)
for pt in video_acked_results['video_acked']:
expt_id = int(pt['expt_id'])
session = (pt['user'], int(pt['init_id']), pt['channel'], expt_id)
# filter data points by congestion control
expt_config = retrieve_expt_config(expt_id, expt_id_cache,
postgres_cursor)
if cc is not None and expt_config['cc'] != cc:
continue
if session not in d:
sys.stderr.write('Warning: ignored session {}\n'.format(session))
continue
video_ts = int(pt['video_ts'])
if video_ts not in d[session]:
sys.stderr.write('Warning: ignored acked video_ts {} in the '
'session {}\n'.format(video_ts, session))
continue
dsv = d[session][video_ts] # short name
# calculate transmission time
sent_ts = dsv['sent_ts']
acked_ts = try_parsing_time(pt['time'])
dsv['acked_ts'] = acked_ts
dsv['trans_time'] = (acked_ts - sent_ts).total_seconds()
return d
def collect_rebuffer(client_buffer_results, postgres_cursor):
# process InfluxDB data
x = {}
for pt in client_buffer_results['client_buffer']:
expt_id = int(pt['expt_id'])
expt_config = retrieve_expt_config(expt_id, expt_id_cache,
postgres_cursor)
# index x by (abr, cc)
abr_cc = (expt_config['abr'], expt_config['cc'])
if abr_cc not in x:
x[abr_cc] = {}
# index x[abr_cc] by session
session = (pt['user'], int(pt['init_id']), pt['channel'],
int(pt['expt_id']))
if session not in x[abr_cc]:
x[abr_cc][session] = {}
x[abr_cc][session]['min_play_time'] = None
x[abr_cc][session]['max_play_time'] = None
x[abr_cc][session]['min_cum_rebuf'] = None
x[abr_cc][session]['max_cum_rebuf'] = None
y = x[abr_cc][session] # short name
ts = try_parsing_time(pt['time'])
cum_rebuf = float(pt['cum_rebuf'])
if pt['event'] == 'startup':
y['min_play_time'] = ts
y['min_cum_rebuf'] = cum_rebuf
if y['max_play_time'] is None or ts > y['max_play_time']:
y['max_play_time'] = ts
if y['max_cum_rebuf'] is None or cum_rebuf > y['max_cum_rebuf']:
y['max_cum_rebuf'] = cum_rebuf
# calculate rebuffer rate
rebuffer = {}
total_play = {}
total_rebuf = {}
for abr_cc in x:
abr_cc_play = 0
abr_cc_rebuf = 0
for session in x[abr_cc]:
y = x[abr_cc][session] # short name
if y['min_play_time'] is None or y['min_cum_rebuf'] is None:
continue
sess_play = (y['max_play_time'] -
y['min_play_time']).total_seconds()
sess_rebuf = y['max_cum_rebuf'] - y['min_cum_rebuf']
# exclude too short sessions
if sess_play < 2:
continue
# TODO: identify and ignore outliers
if sess_rebuf / sess_play > 0.5:
continue
abr_cc_play += sess_play
abr_cc_rebuf += sess_rebuf
if abr_cc_play == 0:
sys.exit('Error: {}: total play time is 0'.format(abr_cc))
total_play[abr_cc] = abr_cc_play
total_rebuf[abr_cc] = abr_cc_rebuf
rebuf_rate = abr_cc_rebuf / abr_cc_play
rebuffer[abr_cc] = rebuf_rate * 100
return rebuffer, total_play, total_rebuf
def collect_rebuffer(client_buffer_results, postgres_cursor):
# process InfluxDB data
last_ts = {}
last_cum_rebuf = {}
outlier_time = {}
excluded_sessions = {}
x = {}
for pt in client_buffer_results['client_buffer']:
expt_id = int(pt['expt_id'])
expt_config = retrieve_expt_config(expt_id, expt_id_cache,
postgres_cursor)
# index x by (abr, cc)
abr_cc = (expt_config['abr'], expt_config['cc'])
if abr_cc not in x:
x[abr_cc] = {}
# index x[abr_cc] by session
session = (pt['user'], int(pt['init_id']), pt['channel'],
int(pt['expt_id']))
if session not in last_ts:
last_ts[session] = None
if session not in last_cum_rebuf:
last_cum_rebuf[session] = None
if session not in outlier_time:
outlier_time[session] = None
if session not in x[abr_cc]:
if session in excluded_sessions:
# ignore sessions that were intentionally removed from x[abr_cc]
continue
x[abr_cc][session] = {}
x[abr_cc][session]['min_play_time'] = None
x[abr_cc][session]['max_play_time'] = None
x[abr_cc][session]['min_cum_rebuf'] = None
x[abr_cc][session]['max_cum_rebuf'] = None
y = x[abr_cc][session] # short name
ts = try_parsing_time(pt['time'])
buf = float(pt['buffer'])
cum_rebuf = float(pt['cum_rebuf'])
# verify that time is basically continuous in the same session
if last_ts[session] is not None:
diff = (ts - last_ts[session]).total_seconds()
if diff > 60: # a new but different session should be ignored
continue
last_ts[session] = ts
# identify outliers: exclude the session if there is one-minute long
# duration when buffer is lower than 1 second
if buf > 1:
outlier_time[session] = None
else:
if outlier_time[session] is None:
outlier_time[session] = ts
else:
diff = (ts - outlier_time[session]).total_seconds()
if diff > 30:
print('Outlier session', abr_cc, session)
del x[abr_cc][session]
excluded_sessions[session] = True
continue
# identify stalls caused by slow video decoding
if last_cum_rebuf[session] is not None:
if buf > 5 and cum_rebuf > last_cum_rebuf[session] + 0.25:
# should not have stalls
print('Decoding stalls', abr_cc, session)
del x[abr_cc][session]
excluded_sessions[session] = True
continue
last_cum_rebuf[session] = cum_rebuf
if pt['event'] == 'startup':
y['min_play_time'] = ts
y['min_cum_rebuf'] = cum_rebuf
if y['max_play_time'] is None or ts > y['max_play_time']:
y['max_play_time'] = ts
if y['max_cum_rebuf'] is None or cum_rebuf > y['max_cum_rebuf']:
y['max_cum_rebuf'] = cum_rebuf
# calculate rebuffer rate
rebuffer = {}
total_play = {}
total_rebuf = {}
for abr_cc in x:
abr_cc_play = 0
abr_cc_rebuf = 0
for session in x[abr_cc]:
y = x[abr_cc][session] # short name
if y['min_play_time'] is None or y['min_cum_rebuf'] is None:
continue
sess_play = (y['max_play_time'] -
y['min_play_time']).total_seconds()
sess_rebuf = y['max_cum_rebuf'] - y['min_cum_rebuf']
# exclude short sessions
if sess_play < 5:
continue
abr_cc_play += sess_play
abr_cc_rebuf += sess_rebuf
if abr_cc_play == 0:
sys.exit('Error: {}: total play time is 0'.format(abr_cc))
total_play[abr_cc] = abr_cc_play
total_rebuf[abr_cc] = abr_cc_rebuf
rebuf_rate = abr_cc_rebuf / abr_cc_play
rebuffer[abr_cc] = rebuf_rate * 100
return rebuffer, total_play, total_rebuf
def collect_buffer_data(client_buffer_results):
d = {} # indexed by session
excluded_sessions = {}
last_ts = {}
last_buf = {}
last_cum_rebuf = {}
last_low_buf = {}
for pt in client_buffer_results['client_buffer']:
session = (pt['user'], int(pt['init_id']), int(pt['expt_id']))
if session in excluded_sessions:
continue
if session not in d:
d[session] = {}
d[session]['min_play_time'] = None
d[session]['max_play_time'] = None
d[session]['min_cum_rebuf'] = None
d[session]['max_cum_rebuf'] = None
ds = d[session] # short name
if session not in last_ts:
last_ts[session] = None
if session not in last_buf:
last_buf[session] = None
if session not in last_cum_rebuf:
last_cum_rebuf[session] = None
if session not in last_low_buf:
last_low_buf[session] = None
ts = try_parsing_time(pt['time'])
buf = float(pt['buffer'])
cum_rebuf = float(pt['cum_rebuf'])
# update d[session]
if pt['event'] == 'startup':
ds['min_play_time'] = ts
ds['min_cum_rebuf'] = cum_rebuf
if ds['min_play_time'] is None or ds['min_cum_rebuf'] is None:
# wait until 'startup' is found
continue
if ds['max_play_time'] is None or ts > ds['max_play_time']:
ds['max_play_time'] = ts
if ds['max_cum_rebuf'] is None or cum_rebuf > ds['max_cum_rebuf']:
ds['max_cum_rebuf'] = cum_rebuf
# verify that time is basically successive in the same session
if last_ts[session] is not None:
diff = (ts - last_ts[session]).total_seconds()
if diff > 60: # ambiguous / suspicious session
print('Ambiguous session', session)
excluded_sessions[session] = True
continue
# identify outliers: exclude the sessions if there is a long rebuffer?
if last_low_buf[session] is not None:
diff = (ts - last_low_buf[session]).total_seconds()
if diff > 30:
print('Outlier session', session)
excluded_sessions[session] = True
continue
# identify stalls caused by slow video decoding
if last_buf[session] is not None and last_cum_rebuf[
session] is not None:
if (buf > 5 and last_buf[session] > 5
and cum_rebuf > last_cum_rebuf[session] + 0.25):
print('Decoding stalls', session)
excluded_sessions[session] = True
continue
# update last_XXX
last_ts[session] = ts
last_buf[session] = buf
last_cum_rebuf[session] = cum_rebuf
if buf > 0.1:
last_low_buf[session] = None
else:
if last_low_buf[session] is None:
last_low_buf[session] = ts
ret = {} # indexed by session
# second pass to exclude short sessions
for session in d:
if session in excluded_sessions:
continue
ds = d[session]
if ds['min_play_time'] is None or ds['min_cum_rebuf'] is None:
# no 'startup' is found
print('No startup found', session)
continue
sess_play = (ds['max_play_time'] - ds['min_play_time']).total_seconds()
# exclude short sessions
if sess_play < 5:
continue
sess_rebuf = ds['max_cum_rebuf'] - ds['min_cum_rebuf']
if sess_rebuf > 300:
print('Warning: bad session (rebuffer > 5min)', session)
if session not in ret:
ret[session] = {}
ret[session]['play'] = sess_play
ret[session]['rebuf'] = sess_rebuf
ret[session]['startup'] = ds['min_cum_rebuf']
print('Valid session count', len(ret))
return ret