Horizon 2020
European Union Funding
for Research & Innovation
Adaptive Schedulers for Deadline-Constrained
Content Upload from Mobile Multihomed
Vehicles
Ali Safari Khatouni
Marco Ajmone Marsan
Marco Mellia
Reza Rejaie (U. Oregon)
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Outline:
Motivation
Problem description
Adaptive algorithm
Experiment setup
Results
Conclusions and future works
2
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Motivation
Video on demand
with deadline
3
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Motivation
How security operator in central control station can
access the portion of the on-board videos which refers
to the period of time of the accident?
Possible Solutions…
Upload videos on deposit Deadlines order of days
Uploading all video for all time Too expensive
Upload on Demand, with requests that must be
satisfied within a deadline
4
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Problem description
Which interface(s) to use?
When to upload data?
At which rate to upload?
Scheduling problem
The Goal is :
Minimizing the cost for uploading requested
videos within the deadline
The cost can be related to money, or to system
cost, or to energy consumption, or to …
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
5
How to solve the problem!
Having FULL knowledge of the network in the
Model by minimum cost flow problem
future
Solved By Centralized Solver
NOT REALISTIC ASSUMPTION!
A. Safari Khatouni, M. Ajmone Marsan, M. Mellia, Delay Tolerant Video Upload from
Public Vehicles, Smart Cities and Urban Computing (SmartCity 2016), April 11, 2016
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
6
New approach
– Assume NO knowledge of the available capacity
– Online and dynamic scheduling:
• Which interface to use to send
• How much data
• While estimating expected rate from current rate
– Goals:
• Minimize the total cost
• Minimize the chance to miss the deadline
7
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Cheap interfaces
(Cost,Rate)
Videos
V1
(Cost,Rate)
(Cost,Rate)
(Cost,Rate)
Expensive interfaces
Source
Sink
(Cost,Rate)
(Cost,Rate)
V2
(Cost,Rate)
8
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Cheap interfaces
Cheap interfaces
Cheap interfaces
β times
more than
expected
Video 1
Video 2
Time
Expensive interfaces
𝑻𝟎
𝑻𝟎 + ∆𝑻
Expensive interfaces
𝑻𝟎 +𝟐∆𝑻
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
9
Algorithm overview
•
•
•
•
Ci,j = EWMA of expected rate for interface i at slot j
Ri,j = actual upload rate for interface i at slot j
Vj = volume of data to be transmitted in slot j
B0 = required average upload rate (file size/deadline)
–
Set Bi = B0 for each time slot
• Main loop:
while(time < deadline or uploaded all data)
1.
Push Vj = BixDT data to buffers
1.
2.
2.
3.
Sort interface by cost, and
Split data on each interface based on expected rate Ci,j so that Si Ci,j=Vj
(simulate upload data base on the collected traces)
Ci,j+1=aCi,j + (1-a)Ri,j # Update the EWMA of the expected rate
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Exponentially Weighted Moving Average (EWMA)
10
Algorithm at a glance
B0 = Min video expected rate to meet
Total uploaded
the deadline
Conservative
data
V
Aggressive
Missing
deadline
t1
We are
late
t2
t3
t4 Deadline
Time
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
11
Adaptive approach
1. Few parameters to tune
1. DT : the duration of the time slot
2. α : the parameter of the EWMA bandwidth
estimator
3. β : the optimism coefficient
12
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Simulation setup
1. Collecting mobile traces with iperf2 in the upload
direction for 600 seconds – more than 50 traces
1. Different technologies (3G, 4G, WiFi)
2. Multiple operators
3. Normal operating condition
Shows how unpredictable is the upload rate
Evaluate the algorithm in realistic setup
Trace driven simulation
13
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Traces are available on : http://tstat.polito.it/traces-MBB-speedtest.shtml
Simulation setup
Simulation Scenario
2 Videos (video size = 62.5 MB)
3 Interfaces (WiFi, 3G, 4G) with cost (2,4,8)
Deadlines [100,600] periods
200 repetitions
At each repetition pick a random trace and choose a random
starting point for the selected trace
Comparison with optimal/centralized solution
Average total upload cost
Percentage of completed upload (over 200 repetitions)
14
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Parameter setting
1. DT = {1,2,5,10} : the duration of the time slot
2. α = {0.1, 0.3, 0.5, 0.7, 0.8, 0.9} : the parameter of the EWMA
bandwidth estimator
3. β = [0,30] : the optimism coefficient
15
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Parameter setting
Pushing more than actual rate
causes high rate of unsuccessful
scheduling
Chosen value to have high
percentage of completed upload
and low cost
16
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Impact of deadline
How to improve it!
Online scheduling is
20-40% worse than
optimal/centralized
solution
17
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Hybrid algorithm
Exploit advantage of both recovery policies
At the first part using conservative policy to get lower
cost
At the second part using aggressive to avoid
uncomplete upload
Where to switch from one policy to another ?
V
Deadline
Conservative
Aggressive
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
18
Result
30% more completed upload
than conservative
12% higher cost with respect to
optimal solution
19
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Conclusions and future work
Propose adaptive scheduler to upload videos with
deadline with NO knowledge of upload rate
Our solution has 12% higher cost than the FULL
knowledge optimal/centralized solution
Using simulation based optimization to solve
parameter setting in real time
Demonstrate and test the whole system in the
MONROE platform
20
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Horizon 2020
European Union Funding
for Research & Innovation
THANK YOU!
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Horizon 2020
European Union Funding
for Research & Innovation
BACKUP SLIDES
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
CDF of traces
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Recovery policy
Do we meet the deadline or not ?
• If we are behind schedule
– recover by scheduling some extra data extraB
• Different policies to recover
– Aggressive: In the next slot => Bi+1=B0 + extraB
– Conservative: in all future slots => Bi+j=Bi + extraB/n
#n=remaining slots
24
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
Results
Greedy in Cost with assumption
of FULL KONWLEDGE of network
as a baseline to COMPARE the
scheduler performance
25
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
How content upload via an interface
3G β = 0
3G β = 5
MEASURING MOBILE BROADBAND NETWORKS IN EUROPE
© Copyright 2026 Paperzz