Data Aggregation in Wireless Mesh Networks A MAC Layer Perspective Romit Roy Choudhury Summer Intern - PowerPoint PPT Presentation

1 / 43
About This Presentation
Title:

Data Aggregation in Wireless Mesh Networks A MAC Layer Perspective Romit Roy Choudhury Summer Intern

Description:

Flights. WLAN. 5. The Mesh Network Vision. Wireless backhaul network of access points (APs) ... Easier to implement (important for cheap routers) ... – PowerPoint PPT presentation

Number of Views:328
Avg rating:3.0/5.0
Slides: 44
Provided by: romitroyc
Category:

less

Transcript and Presenter's Notes

Title: Data Aggregation in Wireless Mesh Networks A MAC Layer Perspective Romit Roy Choudhury Summer Intern


1
Data Aggregation in Wireless Mesh Networks A MAC
Layer PerspectiveRomit Roy ChoudhurySummer
Intern 2005 (Motorola Labs)PhD Student, Computer
Science, UIUCJoint work withYe Chen, Mike
Baker, Steve Emeott (manager)

2
Motivating the Mesh
  • Wireless networks gaining immense popularity
  • Low deployment / maintenance cost
  • Ubiquity, mobility attractive features
  • Data rates on the rise with fast improving PHY
    layer research
  • Popularity bottleneck ? Connectivity
  • Hot-spots require users to come close to it
  • Ideally, the network should move close to the
    (mobile) user
  • Clear demand for extending the hot-spot coverage
  • Email checking during office commute
  • WLAN phones offloading cellular infrastructure
  • Community wireless, Public safety, meshcast, etc.

3
Why different from Cellular
  • Deploy dense overlapping hot-spots and
    inter-connect them with wires
  • Connectivity ensured
  • However, several issues
  • Wiring / deployment cost shown to be huge
    BahlMicrosoft04
  • Deployment a slow process
  • Higher frequency ? higher data rates ? lower
    range ? deployment not scalable
  • Etc. etc etc.
  • Wireless multihop ? cheap, quick, and connected
  • Is that the way to go ??

4
The Macro View
Roads
Electricity
Applications Ubiquity Money
Connected Multi-hop
Wireless Grid
Mesh
Connected Single hop
Cellular
Flights
Vending Machines
WLAN
Single hop
Connectivity Capacity
5
The Mesh Network Vision
  • Wireless backhaul network of access points (APs)
  • Mobile clients (cell phone, laptop, PDA users)

6
Several Research Challenges
  • PHY layer
  • Increase capacity Beamforming, MIMO,
    Modulation, Coding
  • MAC layer
  • Utilize capacity Channel access, Signaling,
    Aggregation, Fairness
  • Exploit characteristics of mesh networks known
    and stationarity topology, persistent and
    predictable traffic, multihop flows
  • Network Transport layer
  • Add connectivity, reliability, robustness, and
    security
  • Application layer
  • Video meshcast, P2P video, public safety, traffic
    information, etc.

7
Problem Definition
  • Improving channel utilization of IEEE 802.11
  • in view of its application in mesh networks.
  • Focus on
  • ? Analysis of data aggregation techniques
  • ? Optimizations to the medium access
    protocol

8
This Talk
  • Data Aggregation
  • Overview and Related work
  • Analysis of Data Aggregation -- Upper bounds and
    Expected case
  • Recommendations on Data Aggregation
  • MAC protocol optimizations
  • Implicit Acknowledgment Hybrid Channel Access
  • Scope for future work
  • Distributed TDMA scheduling (designed for mesh
    networks)
  • Conclusion

9
  • Data Aggregation

10
Recall IEEE 802.11 MAC
d
t
Channel Utilization d / t (assuming success)
RTS
CTS
Data
ACK
Backoff
11
Recent Proposals
  • Recent proposals to 802.11 e/n include data
    aggregation

12
Recent Proposals
  • Recent proposals to 802.11e include data
    aggregation

Several variations in data aggregation, each
leading to different performance tradeoffs
13
Related Work
  • Choi proposed multiple MPDUs followed by single
    ACK
  • ACK can be in subsequent TxOP
  • Jain proposed congestion-adaptive aggregation
  • Aggregation increased when congestion high in the
    network
  • Sadeghi proposed Opportunistic Media Access
  • Back-to-back packets transmitted when channel
    quality is good
  • Trades off short term fairness, but provides long
    term
  • Kanodia proposed Multi-channel opportunistic
    MAC
  • Aggregate on channels that have better channel
    quality
  • Dynamically find such channels

14
Related Work
  • Makhlouf evaluates bounds on aggregation in
    802.11n
  • Analysis considers the upper bounds of achievable
    throughput
  • Assumes error free channel model
  • Previous study interesting, but more to be done
  • Fuller analysis, channel errors, retransmit
    schemes, fairness issues, etc.
  • This study attempts fuller analysis of data
    aggregation
  • Considers the upper bound as well as the average
    case analysis
  • Considers channel errors
  • Considers several retransmission schemes when
    subset of aggregated frames are lost during
    communication

15
Upper Bounds of Aggregation
  • Performance envelope available from the error
    free channel
  • Utilization (u) expressed as ratio of transmit
    time to total time of single dialog (similar to
    Makhlouf)
  • where N of agg. Frames,
  • tp time to transmit payload,
  • B backoff overhead,
  • R channel data rate,
  • C1 rate dependent overhhead,
  • C2 rate independent overhead.

16
Average Case
  • Channel errors can reduce channel utilization
  • Retransmission schemes affect performance
  • Two retransmission schemes considered
  • Block retransmission entire aggregated
    super-frame retransmitted
  • Redundant transmissions reduce channel
    utilization
  • Easier to implement (important for cheap routers)
  • Selective retransmission only corrupted packets
    retransmitted
  • Efficient in terms of performance (throughput and
    delay)
  • Difficult to implement at MAC firmware, higher
    turnaround time

retransmit
1
2
3
4
5
6
7
8
1
2
3
4
5
6
7
8
17
Block Retransmission
  • Assuming i.i.d channel errors for each frame
  • We calculate the average time for transmitting
    super-frame
  • Let T be the R.V denoting the number of slots to
    transmit superframe
  • PT i p(i 1) (1 p)
  • Every retransmission associated with control
    (signaling) overheads
  • Thus, channel utilization expressed as
  • Treating super-frame as single large packet,

18
Utilization under Block Retransmit
  • Higher aggregation better when packet error low
  • At higher channel error, optimal aggregation
    region exists
  • Optimal region a function of specific parameters

19
Selective Acknowledgment / Retransmission
  • Assuming i.i.d channel errors
  • We determine time to transmit all of N packets
    successfully
  • Let t be R.V denoting the time to transmit all
    the packets successfully
  • In general,

20
Selective Acknowledgment / Retransmission
  • Physically interpreting the expression, we have
  • Calculating the expected value of t, we have
  • Substituting from earlier expressions,

t slots
21
Selective Acknowledgment / Retransmission
  • Physically interpreting the expression, we have
  • Thus we have channel utilization as

t slots
22
Benefits with Selective ACK
  • Selective retransmission exhibits clear
    improvement
  • Tradeoff against higher implementation complexity

23
Aggregation in 802.11n
  • IEEE 802.11n ? proposals for high throughput mesh
    network
  • Several MAC layer recommendations on aggregation
  • We utilize our model to analyse recommended
    options
  • Inter-frame spaces reduced to zero (ZIFS)
  • Aggregating MAC Headers (AMPDU)
  • Aggregating PHY Headers (APPDU)
  • We assimilate the best of all worlds
  • Propose combination of ZIFSAMPDUAPPDUSelectiveA
    CK

24
Zero Inter-Frame Spacing (ZIFS)
  • Updating control overhead in our expression with
    ZIFS
  • Plugging into utilization expression

RTS
CTS
Data
Data
Data
Ack
SIFS
ZIFS
25
Aggregating MAC Header
  • MAC headers compressed reduce control overhead
  • However, all frames must be directed to single
    receiver
  • Moreover, sequence numbers cannot be included in
    each frame, thus block retransmission is mandatory

Frame Frame2 Frame 3
Per-frame Header
Common Header
26
Aggregating PHY Header
  • PHY headers can be compressed, without MAC
    compression
  • Small PHY headers ? minor benefits
  • However, MAC sequence numbers make selective
    retransmission possible

Frame Frame2 Frame 3
Per-frame Header
Common PHY Header
27
Combined Aggregation
  • Best of all worlds ? ZIFS AMPDU APPDU Sel
    ACK
  • However, sequence number of each frame necessary
    for selective retransmission
  • Propose bit vector in compressed MAC header
  • E.g., S, 1 0 0 1 1 0 1 0 1 1 represents the
    sequence numbers of frames starting from S,
    included in the superframe.

28
Relative Improvements
  • Normalized comparison of shows
    encouraging benefits
  • With AMPDU, benefits of aggregation reduces
    beyond some value of N
  • With combined scheme, monotonic improvement
    possible

29
Further Improvements ?
  • Data aggregation amortizes overhead over many
    packets
  • However does not reduce the overheads
    significantly
  • Trades off fairness
  • We ask Can we reduce the overhead in the first
    place? without affecting fairness
  • Overheads include RTS/CTS/ACK packets, backoff,
    etc.
  • Overheads consume more than 50 of the channel
    time
  • Studies toward this direction led to new ideas
  • Not enough time to discuss them in detail
  • Brief overview presented in the remaining few
    slides

30
  • Enhancing the MAC Protocol

31
Idea (1) Implicit ACKnowledgment
  • APs on backhaul networks typically backlogged
    with traffic
  • Persistent traffic ? possibility of optimzation
  • We propose an implicit ACK optimization
  • Idea is to piggyback the CTS with ACK for
    previous dialog

802.11
Gain
Implicit ACK
32
Idea (2) Receiver-initiated Polling
  • Implicit ACK eliminates ACK overhead, but RTS,CTS
    remain
  • To further remove overhead, we propose a Hybrid
    Channel Access mechanism
  • A transmitter (T) initiates dialog to receiver
    (R) using RTS/CTS handshake
  • At end of every burst, T intimates receiver (R)
    if it has more packets
  • R performs subsequent channel contention, invites
    T to transmit (via poll) a reversal of role
  • Poll piggybacked with implicit ACK (from
    optimization 1)
  • When T has no more packets to send, R transmits
    explicit ACK
  • Session is closed

33
Hybrid Channel Access
  • The optimization timeline

802.11
Hybrid Channel Access
Implicit ACK
T
R
T
R
T
R
RTS
RTS
RTS
CTS
CTS
CTS
Data
Data
Data
Backoff
ACK
Backoff
Backoff
RTS
Poll ACK
CTS ACK
RTS
Data
Data
CTS
Backoff
Data
Backoff
Poll ACK
RTS
ACK
Data
CTS ACK
Backoff
34
Combined with Data Aggregation
T
R
  • Hybrid scheme eliminates RTS and ACK
  • When persistent backlogged traffic, benefits can
    be high
  • i.e., session length longer
  • Receiver initiated channel access addresses
    hidden terminal problems
  • NAVs set by neighbors of R
  • Possible to perform flow control at R, based on
    Rs buffer length and circuit speed
  • Important for cheaper APs

35
Normalized Results
36
  • Issues and Future Work

37
Issues
  • Several Issues tradeoffs appear with
    aggregation
  • Data aggregation associated with short-term
    unfairness
  • Channel error assumed independent actually
    fading causes correlated packet losses
  • Adaptive aggregation (based on SINR) needs to be
    considered
  • Multi-receiver aggregation still remains
    unchartered area of research
  • Implementation aspects need attention in MAC
    optimizations
  • Deconstructing a super-frame to extract corrupted
    packets only can have high turn-around time
  • Compatibility (of piggybacked poll) with legacy
    802.11
  • Hybrid channel access reduces control packets
  • But does not elimiate the overheads of backing off

38
Future Work Attempt to Eliminate Backoff
  • Mesh APs reserve recurring time slots for future
    communication
  • Reservations distributed to neighbor APs (across
    2 hops)
  • Each AP transmits in their chosen time-slots
    without backing off
  • Mobile users remain unaware of AP schedules ?
    compatibility
  • Outside the reserved time-slots, all nodes (AP
    and mobiles) perform CSMA channel access
  • Duration of time-slots can be adaptive to traffic
    requirements
  • In addition, interactive traffic can be supported
    via CSMA

AP time window
Client traffic
39
Conclusion
  • Connectivity and Capacity bottlenecks to
    todays wireless
  • Mesh networks hold promise however, significant
    research needed
  • Data aggregation can aid in achieving high
    throughputs
  • Analytical models designed to capture the
    efficacy of proposals
  • Combined aggregation (using selective ACK and bit
    vector) prove to be a good candidate
  • The complexity of implementation may be worth the
    effort
  • Data aggregation insufficient to extract maximal
    performance
  • RTS/CTS/ACK and backing off overheads remain
    significant
  • Implicit ACK and Hybrid access schemes alleviate
    overheads
  • However, further efforts needed (e.g.,
    distributed TDMA scheduling) to alleviate channel
    wastage in backing off

40
  • Thanks
  • For your patience

41
Todays Focus
Electricity
Applications Ubiquity Money
Connected Multi-hop
Wireless Grid
Research necessary at PHY, MAC, Network,
Applications, etc.
Mesh
Connected Single hop
Cellular
Ad Hoc WLAN
Single hop
WLAN
Connectivity Capacity
42
Analytical Model
  • Analysis of proposed scheme captures improvement
    in channel utilization(u)
  • Packet errors assumed to be i.i.d
  • where
  • M number of bursts in data sessions
  • N number of data packets aggregated in each
    burst
  • ß backoff duration for each data burst
  • d time for transmitting a single packet

43
Conclusion
  • 802.11 currently most popular for wireless
    networking
  • May not be suitable for future multihop networks
    of high bandwidth
  • Control overhead of 802.11 does not scale with
    higher PHY rates
  • Backing off mechanism a significant bottleneck
  • Internship aim ? Improve 802.11 for next
    generation network
  • Exploit characteristics of mesh networks (static
    topology, traffic, etc.)
  • Solution space includes 3 key ideas
  • Implicit ACKnowledgment (piggybacking scheme)
  • Hybrid channel access (Requiring Tx-Rx-based
    channel access)
  • Distributed TDMA-based scheduling for wireless
    backhaul
  • Evaluations appear to be encouraging ? More work
    required
Write a Comment
User Comments (0)
About PowerShow.com