Post on 24-Oct-2019
Magus: Minimizing Cellular Service Disruption during Network Upgrades
Xing Xu*, Ioannis Broustis#, Zihui Ge#, Ramesh Govindan*,
Ajay Mahimkar#, N.K. Shankaranarayanan#, Jia Wang#
*University of Southern California, #AT&T Labs-Research
{xingx,ramesh}@usc.edu
{broustis, gezihui, mahimkar, shankar, jiawang}@research.att.com
December 3, 2015
ACM CoNEXT 2015, Heidelberg, Germany
1
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
2
Talk Outline
• Problem statement
• Solution strategies
• Validation using LTE Testbed
• Proposed solution: Magus
• Evaluation using operational network data
• Conclusion
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
3
Importance of Good Cellular Services
Smartphones & other mobile devicesplay a central role in our lives
• Communications
• Social networks
• Web-browsing
What provides the connectivity?
• Cellular network!
Cellular networks are constantly upgraded to meetneeds of users
• e.g. software, hardware, power systems
• May take cellular base station offline
• Potential for temporary service disruption
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
4
Service Disruption due to Planned Upgrades
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
5
Plan to take offline
Service Disruption due to Planned Upgrades
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
6
Service Disruption due to Planned Upgrades
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
7
Some users re-attachSome lose service
Service Disruption due to Planned Upgrades
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
8
OverloadedLost service
Service Disruption due to Planned Upgrades
• Network configuration is suboptimal
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
9
OverloadedLost service
Service Disruption due to Planned Upgrades
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
10
Users migrate
Service Disruption due to Planned Upgrades
• Network configuration is suboptimal
• Re-configure neighbors to minimize disruption
• Increase coverage
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
11
Problem Statement
• Planned upgrades which take a base station offline can potentially degrade or disrupt service
• Neighboring base stations can be reconfigured to minimize service disruption
• What are the reconfiguration solution strategies?
• Can we take proactive action, even before a base station is taken offline?
• What should we optimize?
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
12
Reconfiguration Solution Strategies • Reconfiguration type• Model-based: Do what-if analysis for different configurations using a cellular
system model, find optimum, then make network tuning adjustments
• Feedback-based: Make network tuning adjustments, measure performance, and then iterate.
Magus
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
13
Reconfiguration Solution Strategies
• SON (Self-Organizing Networks): Feedback-based, Reactive reconfiguration
• Magus: Model-based, Proactive reconfiguration
Magus
SON
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
14
Comparison of Different Solutions:Benefits of Proactive Reconfiguration
Uti
lity
-15 -10 -5 0 5 10 15
Proactive (Magus) Reactive No Tuning
Upgrade
Proactive Reconfiguration
Reactive Reconfiguration
TimeWe have chosen a Utility Function that balances total sector performance with fair treatment of disadvantaged users
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
15
Exploring Reconfiguration using LTE Testbed
LTE testbed
• Indoor Office deployment
• 5 Cavium eNodeBs
• 10 Sierra Wireless UEs
• Aricent EPC
• 2.6 GHz Band 7 FDD(experimental license)
• Remotely controllable
Case study
• 3 eNodeBs serving 5 UEs
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
16
Exploring Reconfiguration using LTE Testbed
Taking down this cell
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
17
Exploring Reconfiguration using LTE Testbed
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
18
MAGUS: Solution Design
Challenges Magus Approach
What can we reconfigure to recover coverage in a sector
Tune Antenna Tilt and Transmission Power of neighboring base stations
How can we take proactive action? Predictive model for coverage and capacity, for different power & tilt
How do we get an accurate model? Use operational network data for sector coverage, configuration and user distribution
What is a suitable metric for overall service quality?
Use sum log rate metric which balances performance and fairness
How do we search the solution space efficiently?
Use a heuristic algorithm
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
19
Operational Data for Predictive Model
Path loss data
• Operational data for path loss in cellular network
• Enables computation of channel quality (SINR) and estimate of rate
Base Station information
• Location
• Transmit power
• Antenna pattern & tilt
UE distribution
• Distribution of users across grids
Path loss change after antenna uptilt
Path loss example: Sector facing north-west
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
20
Predictive Model – Grid-Level Analysis
Grid information
• Received Powers
Rpa , RPb , RPc …
• Serving eNodeB
argmax(RPs) = a
• SINR
SINR=RPa/(RPb + RPc + N)
eNodeB information
• Grids served
• Users served
g1+g2+… = N
• Throughput
Tmax = f(SINR)
T=Tmax / N
g1 g2 g3 …
At grid:- Rpa , RPb , RPc
- serving eNB- SINR
At eNodeB:- covered grids- # of UEs
(c)
(b)
(a)
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
21
Metrics: Overall Service Quality & Mitigation of Service Disruption
Overall Service Quality
• Maximize rate?
• Maximize coverage?
• Balance of rate and coverage (fairness)
Metric for Utility:
• Utility 𝑈 = 𝑘 log 𝑟𝑘 , where 𝑟𝑘 is rate enjoyed by user k.
• Similar to commonly deployed proportional fair scheduler
• Balances total sector performance with fair treatment of disadvantaged users
Metric for measuring value of Magus approach
• % of Lost Utility that is Recovered
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
22
Performance Improvement Measured withRecovery Ratio
Before
Upgrade
Magus
Recovery ratio = % of Lost Utility which is Recovered
= 𝑈𝑀𝑎𝑔𝑢𝑠 − 𝑈𝑢𝑝𝑔𝑟𝑎𝑑𝑒 𝑈𝑏𝑒𝑓𝑜𝑟𝑒 − 𝑈𝑢𝑝𝑔𝑟𝑎𝑑𝑒
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
23
Evaluation on Operational Data• We evaluated Magus on 3 different markets in different parts of the country
• For each market, we evaluated 3 different areas: Rural, Suburban, Urban
• For each market & area, we evaluated 3 upgrade scenarios
• Offline sectors: 1 central sector, 3 co-located sectors, 4 separated sectors
• Reconfiguration tuning options: Power-only. Tilt-only, Joint power & tilt
Each pixel represents a grid (100m x 100m area)Grids served by the same sector are painted in the same color
~150~50
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
24
Evaluation – Recovery Ratio
• Tilt only tuning < power only tuning < joint tuning
• Best recovery performance in suburban areas
• Urban: interference-limited, overlapping sectors, limited tuning potential
• Rural: large cell radius, difficult for neighbors to cover affected area
• Suburban: sweet spot, neighboring sectors can help to cover the affected area more effectively
Rural Suburban Urban
Power Tuning 18.3% 56.5% 17.1%
Tilt Tuning 8.4% 37.7% 8.8%
Joint Tuning 37.0% 76.4% 20.1%
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
25
Gradual Tuning in Magus
• Take down target cell gradually, doing tuning in stages
• Reduce synchronized handovers
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
26
Evaluation – Benefits of Gradual Tuning• Gradual process for taking sector offline
• Reconfigure with correct sequence of steps
• Reduce simultaneous handovers
Fewer synchronized handovers
Utility never goes below the final configuration
No GradualGradual
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
27
Conclusion and Future Work
Magus: model-based, proactive reconfiguration for planned upgrades
• Predictive model using operational data
• Grid-level analysis:
– SINR at each grid
– Downlink throughput
• Benefits
– Quickly minimize service disruption
– Fewer synchronized handovers
Future work – there is lots !!
• Field trial and deployment of Magus
• Optimization algorithm
• Joint optimization: cross-technology/carriers
• Predictive model for other network events (e.g., radio network congestion)
© 2015 AT&T Intellectual Property. All rights reserved. AT&T, the AT&T logo and all other AT&T marks contained herein are trademarks of AT&T Intellectual Property and/or AT&T affiliated companies. The information contained herein is not an offer, commitment, representation or warranty by AT&T and is subject to change.
28
THANK YOU!