Title: Data Aggregation in Wireless Mesh Networks A MAC Layer Perspective Romit Roy Choudhury Summer Intern
1Data 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)
2Motivating 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.
3Why 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 ??
4The 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
5The Mesh Network Vision
- Wireless backhaul network of access points (APs)
- Mobile clients (cell phone, laptop, PDA users)
6Several 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.
7Problem 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
8This 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 10Recall IEEE 802.11 MAC
d
t
Channel Utilization d / t (assuming success)
RTS
CTS
Data
ACK
Backoff
11Recent Proposals
- Recent proposals to 802.11 e/n include data
aggregation
12Recent Proposals
- Recent proposals to 802.11e include data
aggregation
Several variations in data aggregation, each
leading to different performance tradeoffs
13Related 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
14Related 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
15Upper 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.
16Average 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
17Block 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,
18Utilization 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
19Selective 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,
20Selective Acknowledgment / Retransmission
- Physically interpreting the expression, we have
- Calculating the expected value of t, we have
- Substituting from earlier expressions,
t slots
21Selective Acknowledgment / Retransmission
- Physically interpreting the expression, we have
- Thus we have channel utilization as
t slots
22Benefits with Selective ACK
- Selective retransmission exhibits clear
improvement - Tradeoff against higher implementation complexity
23Aggregation 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
24Zero Inter-Frame Spacing (ZIFS)
- Updating control overhead in our expression with
ZIFS - Plugging into utilization expression
RTS
CTS
Data
Data
Data
Ack
SIFS
ZIFS
25Aggregating 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
26Aggregating 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
27Combined 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.
28Relative Improvements
- Normalized comparison of shows
encouraging benefits - With AMPDU, benefits of aggregation reduces
beyond some value of N - With combined scheme, monotonic improvement
possible
29Further 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
31Idea (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
32Idea (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
33Hybrid 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
34Combined 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
35Normalized Results
36 37Issues
- 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
38Future 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
39Conclusion
- 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 41Todays 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
42Analytical 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
43Conclusion
- 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