Mobile Computing Course Overview
Mobile Computing Course Overview
Instructor: Li Erran Li
(lierranli@cs.columbia.edu)
http://www.cs.columbia.edu/~lierranli/coms
6998-7Spring2014/
3/7/2014: Introduction to Cellular Networks
Review of Previous Lecture
• What are the different approaches of
virtualization?
OS OS OS
Kernel Kernel Kernel
Hypervisor / VMM
Hardware
OS OS OS
Hypervisor / VMM
kernel emulated
Host OS Kernel module devices
Hardware
3/7/14 Cellular Networks and Mobile Computing
Courtesy: Jason Nieh et al. 5
(COMS 6998-10)
Review of Previous Lecture
(Cont’d)
• What approach does Cell use?
• What are the key design choices for Cell’s
extremely low overhead?
RTC / Alarms
Framebuffer
Audio/Video
Cell Radio
Android...
Sensors
Power
Input
GPU
WiFi
•••
Performance MB/s
RiData 2 27 7.9 0.02
flash storage Sandisk 4 23 5.5 0.70
• Mobile flash storage Kingston 4 25 4.9 0.01
classified into speed Wintec 6 25 15.0 0.01
classes based on A-Data 6 30 10.8 0.01
sequential throughput Patriot 10 29 10.5 0.01
Random write performance is PNY 10 29 15.3 0.01
orders of magnitude worse
Consumer-grade SD performance
D-AMPS
PDC 3GPP
A base station
with 3 sectors
(3 cells)
Radio basics
User #1 scheduled
User #2 scheduled
IuCS IuPS
•Real-time radio control Complex, real-
•Radio Resource Management
RNC •Soft handover time RAN
•UP Ciphering
•Header Compression
Iub
•L1
NodeB •HSPA scheduling
Vendor lock-in
due to
3G Radio Access Network
proprietary Iub
features Courtesy: Zoltán Turányi
From 3G to EPC/LTE architecture
Only two data
PS Core Network plane nodes in the Evolved Packet Core (EPC)
typical case.
Gi SGi
CS
GGSN PDN GW Packet Data Network GW
Serving GW
SGW
Gn/Gp
MSC S11
control plane Mobility
SGSN MME Management
Entity
IuCS IuPS Data plane/control
plane split for PS only
RNC better scalability. S1-UP S1-CP
Iub
S5/S8
SGW Serving GW
S11
Mobility
MME Management SGW and PDN GW separate in
Entity
some special cases:
S1-UP S1-CP
• Roaming:
• PDN GW in home network,
• SGW in visited network
• Mobility to another region in a
eNodeB eNodeB – Evolved Node B large network
• Corporate connectivity
LTE Radio
Access Network
Courtesy: Zoltán Turányi
36
Interworking with 3G
SGi
PDN GW HSS
S5 Gn
eNodeB NodeB
PDN GW HSS
S5 Gn
S2
SGW S11 MME SGSN MSC
IuPS IuCS
(cdma2000, WiMax,
WiFi) eNodeB NodeB
UE
PMIP – Proxy Mobile IP
3/7/14 Cellular Networks and Mobile Computing Courtesy: Zoltán Turányi 38
(COMS 6998-10)
Debate of 2006:
GTP vs. PMIP
SGi
PDN GW HSS
GTP
GTP?
S5 GTP Gn
PMIP
PMIP?
S2
PMIP SGW S11 MME SGSN MSC
IuPS IuCS
(cdma2000, WiMax,
WiFi) eNodeB NodeB
SGi
PDN GW HSS
S5 GTP Gn
GTP
eNodeB NodeB
UE
3/7/14 Cellular Networks and Mobile Computing 40
Courtesy: Zoltán Turányi
(COMS 6998-10)
41
PDN GW HSS
S5
PMIP
S2
PMIP SGW S11 MME
S1-U S1-CP
Non-3GPP GTP
Access
(cdma2000, WiMax,
WiFi) eNodeB
• Procedure
5
1. UE measures nearby cells
2. UE sends measurement reports to network 3. 4.
48
PAGING
• UE periodically checks if data is available for it
– Wakes up, (re)selects cell, reads broadcast and the paging
channel
– Exact timing is pseudo-random per UE
› If packet arrives to SGW…
– …it buffers the packet
– …and notifies MME. PDN GW
– MME sends a Paging Request to all eNodeBs
in the TA of the UE SGW MME
– eNodeBs page the UE on its paging slot
locally
– UE responds with a Service Request…
– …eNodeB state is built up…
– …and UE is moved to connected state. Courtesy: Zoltán Turányi
UE 49
Idle mode issues
• Idle mode is a great power-saving feature
– A system-wide feature
– Also saves a lot of RAN resources
• Balancing of TA size is needed
– Too large: many paging messages
– Too small: many TAU messages from UE
– Lot of optimizations: per-UE TA, overlapping TA, etc.
• Connected Idle transitions are costly
– Usually a timeout is used to go to idle
• Not a good fit for chatty packet traffic
• Easy to attack: an IP address range scan wakes up everyone
– Key application design goal: reduce chattyness
• The Phone OS also has responsibility
– However, can be very effective when combined with DRX
Cellular Networks and Mobile Computing
(COMS 6998-10) 50
LTE RRC State Machine
• UE runs radio resource
control (RRC) state
machine
• Two states: IDLE,
CONNECTED
• Discontinuous reception
(DRX): monitor one
subframe per DRX cylce;
receiver sleeps in other
subframes
Tail Time
Delay: 2s Delay: 1.5s
Channel Radio
Power
IDLE Not Almost
allocated zero
CELL_FACH Shared, Low
Tail Time Low Speed
CELL_DCH Dedicated, High
3/7/14
High Speed
Courtesy: Feng Qian Cellular Networks and Mobile Computing
52
(COMS 6998-10)
Why Power Consumptions of RRC States
so different?
• IDLE: procedures based on reception rather
than transmission
– Reception of System Information messages
– Reception of paging messages with a DRX cycle
(may trigger RRC connection establishment)
– Location and routing area updates (requires RRC
connection establishment)
• CONNECTED: need to continuously receive, and
sent whenever there is data
55
MSISDN – Mobile Subscriber ISDN Number
KEY hierarchy
SGi AuC
PDN GW HSS USIM / AuC K
S5
CK, IK AKA procedure
UE / HSS
SGW S11 MME KASME
UE / MME
KNASenc KNASint
S1-U S1-CP
KeNB / NH
UE / eNB
http://msc-generator.sourceforge.net v3.4.18 57
handover
USIM / AuC K
UE source eNB target eNB MME SGW PDN GW
User Data CK, IK
UE / HSS
1: Measurement
KASME
report
UE / MME
2: Handover decision
KNASenc KNASint
3: Handover
Request KeNB / NH
{NH, NCC} UE / eNB
4: Allocate TEID
5: Handover KUPint KUPenc KRRCint KRRCenc
6: handover
command Request Ack
7: SN Status
Transfer
User Data
buffer DL data
• MME pre-calculates NH keys
8: Sync+RRC complete – From KASME and NCC
User Data – NCC: NH Chaining Counter
• 3: Source eNodeB sends
9: Path Switch
Request 10: Modify Bearer {NH, NCC} to target eNodeB
Request
User Data end marker
• Target eNB uses NH for KeNB
• UE also calculates new KeNB
stop fw stop fw
QoS architecture
PDN-GW HSS
• A bearer is a L2 packet transmission S5
channel
SGW S11 MME
– …to a specific external Packet Data Network,
– …using a specific IP address/prefix, S1-U S1-CP
PDN GW PDN GW
Dedicated bearer: bearer with special QoS
Default bearer: rest of traffic with default QoS
SGW MME
eNodeB
Two default bearers
to different APNs PDN – Packet Data Network
UE Courtesy: Zoltán Turányi APN – Access Point Name 61
Why then no QoS today?
(Apart from voice)
5
50 ms
300 ms
10
10
-3
-6
Real Time Gaming
– 9 mandatory,
sharing, progressive video, etc.)
7 Non-GBR Voice,
(NOTE 3) 7 100 ms -3 Video (Live Streaming)
10
• MBR: Max bitrate 300 ms -6 TCP-based (e.g., www, e-mail, chat, ftp, p2p file
10
9 9 sharing, progressive video, etc.)
(NOTE 6)
1. Session setup
No QoS API
App App
Function Rx
SGi
UE
Courtesy: Zoltán Turányi
65
What Is Next?
LTE-A
Rel-11
LTE
Rel-10
Rel-9
3/7/14 Cellular Networks and Mobile Computing 67
(COMS 6998-10) Rel-8
LTE Evolution
• LTE-B
– Work starting fall 2012
• Topics (speculative)
– Device-to-device communication
LTE-C
– Enhancements for machine-to-machine
Rel-14
communication Rel-13
LTE-B
– Green networking: reduce energy use Rel-12
Rel-9
3/7/14 Cellular Networks and Mobile Computing 68
(COMS 6998-10) Rel-8
Outline
Goal of this lecture: understand the basics of
current networks and future directions
• Current Cellular Networks
• What Is Next?
• A Clean-Slate Design: Software-Defined Cellular
Networks
– Radio Access Networks
– Core Networks
– Wide Access Networks
• Conclusion and Future Work
3/7/14 Cellular Networks and Mobile Computing 69
(COMS 6998-10)
A Clean-Slate Design:
Software-Defined Radio Access
Networks
2017
10
15
17.5
2.5
12.5
0
5
7.5
20
-12.5
-15
-10
-5
-2.5
• Poor wireless connectivity if left unaddressed
71
LTE Radio Access Networks
• Goal: high capacity wide-area wireless network
– Dense deployment of small cells
Base Station (BS)
Serving Gateway
Packet Data
User Equipment (UE) Network Gateway
Internet
Serving Gateway
72
access core
Dense and Chaotic Deployments
74
SoftRAN: Big Base Station Abstraction
Big Base Station
Radio Element 1
time
controller
frequency
Radio Element 2 Radio Element 3
time time
time
frequency frequency
75
Radio Resource Allocation
time
76
SoftRAN: SDN Approach to RAN
Coordination :
X2 Interface
BS2 BS4 77
SoftRAN: SDN Approach to RAN
Control Algo Operator Inputs
Network OS
RadioVisor
RE3
RE1 RE5
PHY & MAC PHY & MAC
Radio Element
(RE)
RE2 RE4 78
SoftRAN Architecture Summary
CONTROLLER
3D Resource Grid
Radio Element
79
SoftRAN Architecture: Updates
• Radio element -> controller (updates)
– Flow information (downlink and uplink)
– Channel states (observed by clients)
80
SoftRAN Architecture: Controller Design
• RAN information base (RIB)
– Update and maintain global network view
• Interference map
• Flow records
• Radio resource management
– Given global network view: maximize global utility
– Determine RRM at each radio element
81
SoftRAN Architecture: Radio Element API
82
Refactoring Control Plane
• Controller responsibilities:
- Decisions influencing global network state
• Load balancing
• Interference management
83
SoftRAN Advantages
• Logically centralized control plane:
– Global view on interference and load
• Easier coordination of radio resource management
• Efficient use of wireless resources
– Plug-and-play control algorithms
• Simplified network management
– Smoother handovers
• Better user-experience
84
SoftRAN: Evolving the RAN
• Switching off radio elements based on load
– Energy savings
• Dynamically splitting the network into Big-BSs
– Handover radio elements between Big-BSs
85
Implementation: Modifications
• SoftRAN is incrementally deployable with
current infrastructure
– No modification needed on client-side
– API definitions at base station
• Femto API : Standardized interface between scheduler
and L1 (http://www.smallcellforum.org/resources-
technical-papers)
• Minimal modifications to FemtoAPI required
86
RadioVisor Design
• Slice manager
o Slice configuration, creation,
modification, deletion and multi-slice
operations
3D Resource • Traffic to slice mapping at RadioVisor
Grid Slice and radio elements
Traffic to
Slice Allocation & Manager • 3D resource grid allocation and
Mapping Isolation isolation
RadioVisor o Considers traffic demand, interference
graph and policy
87
Slice Manager
• Slice definition
o Predicates on operator, device, subscriber, app
attributes
o A slice can be all M2M traffic of operator 1
• Slice configuration at data plane and control plane
o PHY and scheduler: narrow band PHY for M2M slice
o Interference management algorithm
• Slice algebra to support flexible slice operations
o Slice merge, split, (un)nest, duplicate
88
Resource Grid Allocation and
Isolation
• Slices present resource Interference
Edge
demands every time window
• Max min fair allocation
• Example
Radio Radio Radio
o Red slice entitles 2/3 and Element 1 Element 2 Element 3
demands 2/3 RE1 only
o Blue slice entitles 1/3 and
Frequency
demand 1/3 RE2 and 1 RE3
89
Conclusion
• Dense deployment calls for central control of radio
resources
• Deployment costs motivate RAN Sharing
• We present the design of RadioVisor
o Enables direct control of per slice radio resources
o Configures per slice PHY and MAC, and
interference management algorithm
o Supports flexible slice definitions and operations
A Clean-Slate Design:
Software-Defined Cellular Core
Networks
Serving Gateway
Packet Data
User Equipment (UE) Network Gateway
Internet
Serving Gateway
access core 92
SoftCell Overview
Simple hardware
+ SoftCell software
Controller
Interne
t
93
SoftCell Design Goal
Fine-grained service policy for diverse app needs
» Video transcoder, content filtering, firewall
» M2M services: fleet tracking, low latency medical
device updates
94
Characteristics of Cellular Core
Networks
1. “North south” traffic pattern
2. Asymmetric edge
3. Traffic initiated from low-bandwidth access
edge
Gateway Edge
Internet
~1 million Users
~10 million flows
~400 Gbps – 2 Tbps
Two Regions
101
Mobile WANs Problems
• Suboptimal routing in large carriers
– Lack of sufficiently close PGW is a major cause of
path inflation
• Lack of support for seamless inter-region
mobility
– Users crossing regions experience service
interruption
• Scalability and reliability
– The sheer amount of traffic and centralized policy
enforcement
• Ill-suited to adapt to the rise of new
applications
– E.g., machine-to-machine
– All users’ outgoing traffic traverses a PGW to the 102
SoftMoW Motivation
Question: How to make the packet core scalable, simple,
and flexible for tens of thousands of base stations and
millions of mobile users?
• Mobile networks should have fully connected core
topology, small logical regions, and more egress points
• Operators should leverage SDN to manage the whole
network with a logically-centralized controller:
– Directs traffic through efficient network paths that
might cross region boundaries
– Handles high amount of intra-region signaling load
from mobile users
– Supports seamless inter-region mobility and
optimizes its performance
– Performs network-wide application-based such as
region optimization
103
SoftMoW Solution
• Hierarchically builds up a network-wide control
plane
– Lies in the family of recursive SDN designs (e.g. XBAR,
ONS’13)
• In each level, abstracts both control and data planes
and exposes a set of “dynamically-defined” logical
components to the control plane of the level above.
– Virtual Base stations (VBS), Gigantic Switches (GS), and
Virtual Middleboxes (VMB)
Latency Union of Sum of
Matrix Coverage capacities
GS VBS
VMB
VBS1
VBS2
VBS2
GSW1 GSW2
VBS3 GSW2
GSW3
VBS3
Merge/Split Move and Split
105
First Level-SoftMoW Architecture
• Replace inflexible and expensive hardware devices (i.e.,
PGW, SGW) with SDN switches
• Perform distributed policy enforcement using middle-box
instances
• Partition the network into independent and dynamic logical
regions Events GS Rules &
• A child controller manages the data plane of each regions
Actions
Local
Bootstrapping Agent A Apps
phase: NIB Child A
based on location E2 E3
and processing Boundary
capabilities of child M 1 M
Region A 6 E4
controllers E1 Region B
I1 M M
M 2 M 3 7 8
M M
M 4 M 5 9 10
Abstraction
Handover update
coordination
graph Child A Child B
E2 E3
Boundary
Parent M 1 M
Region A 6
E1 Region B
E1 E2 E3 E4 I1 M M
M 2 M 3 7 8
M M
I1 2M M GS Rule: M M
M M GSA GSB M 4 M 5 9 10
2M M
Move Border VBS1
Old
Internal Border Border Internal New
Border 109
VBS VBS1 VBS2 VBS Border
1 2
Conclusion
SoftMoW:
• Brings both simplicity and scalability to the
control plane of very large cellular networks
– decouples control and data planes at multiple levels
( focused only on two levels here)
• Makes the deployment and design of network-
wide applications feasible
– E.g., seamless inter-region mobility, time-of-day
handover optimization, region optimization, and
traffic engineering
110
Summary
• Mobile computing depends on cellular
networks
• Cellular network performance still far from
meeting demands of mobile computing
• Cellular network architecture is evolving to
meet demands of mobile computing
– SDN and NFV
• AT&T’s domain 2.0