Slide - Queen`s School of Computing

Lookback Scheduling for Long-Term
Quality-of-Service Over Multiple Cells
QoS and QoE in Wireless Communications/Networks Workshop
(QoS-QoE 2013), 9th IWCMC’13, Cagliari, Italy.
Hatem Abou-zeid*, Stefan Valentin†, Hossam S. Hassanein*,
and Mohamed F. Feteiha*
* Queen’s University, Canada
†Bell Labs, Alcatel-Lucent, Germany
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Introduction: Downlink Scheduling Basics
User Queues
Incoming
user data
from the core
network
Base
Station
User Playback
Buffers
.
.
.
.
.
.
Proportional-Fair Scheduler (PF):

Schedule user with highest
[Channel Condition] ri (t )
[ Average Rate]
Ri (t )


Throughput-fairness balance
Ri(t) computed over a window Tw
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Introduction: Downlink Scheduling
User Queues
Incoming
user data
from the core
network
Base
Station
.
.
.
User Playback
Buffers
.
.
.
Exponential Scheduler (EXP):


Schedule user with highest
1 N
qi (t )  i 1 qi (t )
ri (t )
N
exp
Ri (t )
1 N
1
 qi (t )
N i 1
Idea: when a user queue increases relative to
average queues, the user is prioritized exponentially
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Traditional Scheduling: Short-term QoS Indicators



Traditionally, schedulers employ QoS indicators such as
average rates Ri(t) to provide service guarantees and
fairness.
These indicators are usually computed over a short
duration, typically a few seconds.
Further, QoS indictor information from prior cells is not
transferred to the user’s current cell
The QoS a user receives in one cell will not impact the
future QoS in upcoming cells as
BSs only know the user QoS in their cell
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Motivation for Proposing Long-term QoS
Today’s networks have fluctuating demand:




in different cells
at different times of the day
network traffic is uneven in space and time
Today’s mobile usage involves:




Longer user sessions and more video content
Highly mobile users
users traverse multiple cells during a single session
Users receive variable QoS as they move
throughout the network
5
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Long-Term Multi-cell QoS
Long-term notion of scheduling enables cell
cooperation over time by looking back


Users served poorly in congested cells can be compensated
in future cells
Proposal: BSs monitor and exchange long-term user
QoS

Result: improve long-term user satisfaction
and reduce subscriber churn
6
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Simple Scenario: Achieving Long-term QoS
1
Congested cell
Percentage of Frozen Video
0.9
0.8
Distance [km]
0.7
0.6
0.5
Vacant cell
0.4
0.3
Base Stations
Mobile User 1
Mobile User 2
Static Users
0.2
0.1
0
0
7
0.2
0.4
0.6
Distance [km]
0.8
40
30
Fairness in
Video Quality
(Freezing)
20
10
0
1
User 1
User 2
Without LLS With LLS
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Look-back Scheduling (LLS)
Look-back Scheduling adds long-term QoS indictors into the
scheduling decision.



This means that user QoS is monitored either by the user, or the
network, and reported during handover.
LLS scheduler should also be aware of users immediate resource needs.
LLS design factors:



8
Which utility functions to use for short and long-term QoS
indicators?
How do you combine then for an overall user utility?
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Look-back Scheduling: Proportional Fairness (LL-PF)
Long-term user
throughput over
multiple BS
QoS metrics
from users on
perceived
quality
Compute
Long-term
User
Satisfaction
Long-term
indicators
Combine
Long and
Short-term
Indicators
Final User
Scheduling
Priority
Channel User
Quality Rate
Short-term
indicators
Long-term Look-back PF Scheduler
9
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
LL-PF: Effect of a
Slot Rate Metric: 10th percentile
throughput




Computed over T slots
Indicates slot starvation
level
Will be zero if user is
starved for more than 10%
of the time slots
A high value indicates that
user is served well in the
worst 10% of the time slots
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Look-back Scheduling: Exponential (LL-EXP)
Long-term user
throughput over
multiple BS
QoS metrics
from users on
perceived
quality
Compute
Long-term
User
Satisfaction
Long-term
indicators
Combine
Long and
Short-term
Indicators
Final User
Scheduling
Priority
Channel Queue
Quality Lengths
Short-term
indicators
1 N
i 1 qi (t)
ri (t )
*
N
i  arg max LT exp
x
Ri ( t )
1 N
1
q (t )

i 1 i
N
qi (t ) 
Long-term Look-back EXP Scheduler
11
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Performance Evaluation
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
System Model


19 Cell Network, 1km inter-BS distance
Mobility: Random Waypoint 2.5
Channel:
2




Traffic:



Path-loss: 128.1 + 37.6 log(d)
Slow fading: 8 dB log-normal
Fast fading: i.i.d. Rayleigh



1
0.5
0
Full buffer,
Constant bit-rate for video traffic
-0.5
Metrics:

1.5
Y (Km)

-1
Network throughput
-1.5
Jain’s Fairness Index
10th percentile slot throughput:-2
-2.5
Average freezing
-3

-2
-1
Average ratio of playback time that is frozen for all
0
X (Km)
users
in the
1
network
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
2
3
Fairness Results: LL-PF Scheduling

LL-PF provides long-term fairness over multiple cells, while
simultaneously providing short-term rates depending on the
tuning factor a
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Throughput Results: LL-PF Scheduling

LL-PF network throughput is also higher than PF for values of
a that provide a similar short-term slot rate

Therefore there are gains in both throughput and fairness
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Results: LL-EXP Scheduling

LL-EXP achieves throughput and video freezing gains


The long-term average rate computation allows the scheduler to exploit
user channel opportunistically
Gains increase with load
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Summary




In this paper we introduce the notion of Long-term Look-back
Scheduling (LLS) over multiple cells.
To achieve this we propose that QoS indicators are monitored
during a user session, and incorporated along with traditional shortterm indicators to make the overall scheduling decision
This introduces some signalling during hand-over, where the BS or
the user, should transmit the QoS indictors to the target BS.
We developed two LLS and assessed their performance:
 Proportional fair scheduling with long and short-term user rates
 Exponential scheduling with long and short term QoS indicators
 Long-term user QoS gains were observed in both cases.
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells
Thank You

Questions?

Please feel free to contact us at [email protected]
Lookback Scheduling for Long-Term Quality-of-Service Over Multiple Cells