Wireless Data Networking Research: From Concept to Practice Songwu Lu UCLA
Mar 27, 2015
Wireless Data Networking Research: From Concept to
Practice
Songwu Lu
UCLA
Drivers for Wireless Networking Research
Transport Layer
Network Layer
Link Layer
New Services, Architectures, Requirements
New Wireless Communications Technology
Top
Down Bottom
Up
Key Driver: Wireless Communications
• Many examples of them:– Sector Antenna, antenna arrays, Smart antennas – Adaptive modulation, MIMO, OFDM, UWB, .. – Cognitive radio, software radio, spectrum sharing, channel
management– Multiple radios, device heterogeneity– …
• many orthogonal dimensions– RF spectrum, antenna, data processing, …
• main goal: improve performance in terms of spectral efficiency
• Challenge: How to exploit these new PHY communication capabilities in the protocols?
Root Cause of Problems
two largely disconnected communities• speak different terminologies
– wireless communications:• Symbols, signals• probabilistic terms:
– information theoretic bounds– confidence factor on symbol reception, …
– wireless networking• Packets, bits• deterministic terms
– Correct/wrong binary reception
Root Cause of Problems (2)
Two largely disconnected communities• different methodologies
– wireless communications• solid theoretic foundation on information theory• a set of well known assumptions: noises, interferences, etc.• Theory Design-->Analysis-->prototype in chips-->experiments
– wireless networking• mostly on heuristics • network setting “ad hoc”: no agreed benchmarks/base settings• Heuristic Design-->Simulations--Network Prototype--
>Experiments
Perspective From Wireless Networking
• We are not on the driver’s seat so far– communication has driven the technology so far– we are followers
• No need to be sad– still plenty of space
• the direct communication almost NEVER works in reality at the 1st place!
– other brothers also facing similar situations sometime• Internet: PC/hardware industry• Cellular: mobile phones
Research Life Cycle in Traditional Wireless Networking Researcher
1. wait for new radio communication tech. to come to life
2. be the 1st to design networking solution to it3. not so lucky?
• understand the problem better• check other aspects/components in the system
4. apply the set of tricks in your bag5. claim credit/declare failure
• Experiments!!!!• Positive success: insights learned• Negative failure: lessons learned
Two Design Guidelines
2 most popular design principles used in the research community
1. Adaptation
high-dimension dynamics
2. Coordination
coherent system
Bag of Tricks in Adaptation• Model-referenced design
– Ideal model to capture expected behaviors under idealized situation
• e.g., error-free, static settings– Track the reference model under realistic conditions/scenarios
• Mobility, wireless channel dynamics, …
• Opportunistic design approach– Make each perform under peak conditions– Exploit the system population– Leverage system diversity
• Multiple receivers, multiple devices, multiple applications/flows, …
Bag of Tricks in Coordination
• Cross-Layer design– not integrated design cross layers– information sharing, informed decision at other
layers– …
• Coordination via “indirection”– Adaptation-aware proxy provides indirection: act
as converter/translator
Illustration Case: Rate Adaptation in Wi-Fi
Problem: Adapt transmission rate to channel quality
Sender Receiver
54MbpsSignal is goodSignal becomes weaker
12Mbps
•The 802.11 a/b/g/n standards allow for multiple rates based on adaptive modulation
–802.11b: 4 rate options (1,2,5.5,11Mbps)–802.11a: 8 options (6,9,12,18,24,36,48,54)–802.11g: 12 options (11a set + 11b set)
•unspecified by the IEEE 802.11 standard
As the Lucky, 1st Guy
• Driver: adaptive modulation• Good news: SNR based feedback not there!• Opportunity: packet-level information available• Solution:
– Hypothesis: packet loss indicates channel quality change
– Tricks: • Decrease transmission rate upon severe packet
loss• 10 consecutive successes → increase rate
Rules For Not So Lucky?
Understand the problem better– if a problem is not better understood, it is
probably best not to provide a new solution at all
– no rush for quick solutions• incremental improvement is #1 enemy in
research!
– do not improve on flawed design!!• adding gas into fire
Experiments to Discover (No) Problems
ARF AARF SampleRate Fixed Rate
UDP Goodput (Mbps)
0.65 0.56 0.58 1.46
SenderReceiverHiddenStation
Case: packet collision scenario?
The sender performs worse with Rate Adaptation!
• The sender should not decrease the rate upon collision losses– Decreasing rate increases collisions !
Find Root Cause
Severeloss
DecreaseTxrate
IncreaseTxtime
IncreasecollisionProb.
Fail to handle hidden-station!
Solution?
• Straightforward idea: RTS/CTS
• more thinking: make RTS/CTS adaptive– reduce overhead– infer collision levels
• Performance: ~80% throughput gain
Software
Hardware
PHYfeedba
ck
802.11 MAC
AdaptiveRTS
Loss EstimationRate Selection send
RTSOption
RRAA
Now MIMO Case?
• Driver: 802.11-pre-n MIMO
• Good/Bad News: SNR feedback to some extent– more direct & timely information on channel
quality?– Loss-based design obsolete?
SNR vs Rate vs Throughput
•SNR vs rate vs thruput are non-monotonic in fine grain•main trend can still be correct
•RF Chamber experiments
Solution in MiRA
• using SNR pre-selects a range of rates– determine a rate window [minRate, maxRate].
• Loss-based best rate choice within the window – play old tricks using loss-based design
Experiments on Static Clients: UDP
Gains in blue arrows refer to MiRA vs. Atheros RA
Static Clients Scenario: TCPGains in blue arrows refer to MiRA vs. Atheros RA
Broader View on Well-Known Areas
• look at other systems component the design works with
• illustrative example: Network Coding– hot topics– several papers on top conferences, from groups
@ MIT, Microsoft Research, …
– what can I do?
Network Coding in Reality: Wi-Fi Nets Multicast/broadcast (a XOR
b) @ 6Mbps Base rate without RA Used in COPE, Wi-Fi
broadcast
NC is worse ! Xmit time w/o NC
2L/54 + 2L/24 Xmit time with NC
L/54 +L/24+L/6
Conclusion: NC works but loses without any RA!
AliceBob
a b
a XOR b
54Mbps 24Mbps
Native NC (@ base rate) May NOT gain at all !
NC Gain May Vanish
Simple multicast RA solution: multicast = min
(rate_receiver) NC gain reduces
NC: 25% (4 tx ->3 tx) In the literature
Actual gain (11a): 5% NC tx time: 2*L/6+L/54 No NC: 2*L/54+2*L/6
802.11b: 1/24 (11M&1M) Root cause: NC cannot
exploit rate diversity!
AliceBob
a b
a XOR b
54Mbps 6Mbps
NC gain (@ optimal rate) may reduce in rate diversity case!
My View on New Frontiers
• no need to get squeezed in crowded traditional areas
• bag of tricks grow much slower!
• problem space is wild wide west!
Wireless Networking on a Chip
• 1000s of cores Systems on a Chip
• wired interconnect: latency, physical wiring constraints• High-speed wireless shortcuts
Composable Wireless Networking
• composable & modular from radio to networking
• Radios become dynamically loadable modules– no clear separation of multi-radios – Software Defined Radios platforms
“Green” Wireless Infrastructure
• infrastructure is power hungry – asymmetric design in cellular network
• more complexity @ base stations– from radio communication, to signaling, to higher
layers
• lots of energy-saving proposals @client side– no on the infrastructure
Resilience-Oriented Design
• mostly performance driven for wireless networking so far
• resilience as the 1st principle– not as patches– learn the success from the Internet
• still early to have a nice try
Still Unhappy? Looking Up• New requirements
– Security, privacy, robustness/dependability, distributed management
• New applications and services– MMS, P2P image/video sharing, IP TV streaming, …– (Location-based, context-aware, personalized,
pervasive) services
My View on Pervasive Cloud Computing
• Data stored in the “Cloud”• Data follows you & your devices• Data accessible anywhere• Data can be shared with others
music
preferences
maps
newscontacts
messages
mailing lists
photo
e-mails
calendar
phone numbers
investments
“Anytime, Anywhere, Any device” Data Service
Research Sub-areas
1. Data Center Networking: Improving the Cloud Infrastructure
2. New Services For Mobile Devices– Security: Virus detection– Location-based Service,– social networking,…
3. Better Access for the Client1. Improving Wi-Fi, 3G+, … for user access2. Opportunistic Client-Client Service
Final Words
• Life can be good or bad in wireless networking research– It is more about your choice
• You are part of inventing the artifact for wireless networking
Acknowledgments
• most real work is done by the real heroes in projects:– Students:
• Innaois Yannis, Suk-Bok Lee, Starsky Wong, Hao Yang, Haiyun Luo,…
– Colleagues: • Lixia Zhang, Mario Gerla, Chuanxiong Guo,
Jacky Shen, Yongguang Zhang, Shugong Xu, …