towards a robust protocol stack for diverse wireless networks
DESCRIPTION
Towards a Robust Protocol Stack for Diverse Wireless Networks. Arun Venkataramani (in collaboration with Ming Li, Devesh Agrawal, Deepak Ganesan, Aruna Balasubramanian, Brian Levine, Xiaozheng Tie at UMass Amherst). Cellular. DTN. Mesh. WLAN. MANET. High mobility. Little mobility. - PowerPoint PPT PresentationTRANSCRIPT
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS, A, AMHERST • MHERST • Department of Computer Science Department of Computer Science
Towards a Robust Protocol Stack for Diverse Wireless Networks
Arun Venkataramani(in collaboration with Ming Li, Devesh Agrawal, Deepak Ganesan, Aruna
Balasubramanian, Brian Levine, Xiaozheng Tie at UMass Amherst)
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 2
Wireless network landscape today
Diverse wireless networks coexistAdapt TCP/IP stack in different ways
WLAN Mesh MANET DTN
Little mobility High mobilitySome mobility
Cellular
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 3
Interconnecting diverse networks?
Vision: A simple robust protocol stack to interconnect diverse wireless networks.
DTNWLAN Mesh MANET
Little mobility High mobilitySome mobility
Cellular
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 4
Why does TCP/IP not suffice?
E2E route disruptions cause unavailability
DTNWLAN Mesh MANET
Little mobility High mobilitySome mobility
Cellular
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 5
Why does TCP/IP not suffice?
Gap between what you buy vs. get
DTNWLAN Mesh MANET
Little mobility High mobilitySome mobility
Cellular
Advertised capacity: 11Mbps
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 6
Principle: Design for high uncertainty
Observe what works at an extreme point in the design space--always-partitioned DTNs--for insights into a robust design.
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 7
Outline
Motivation
Block transport
Replication routing
Research challenges
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 8
E2E Feedback
1. E2E Transport
E2E rate control is error-prone
E2E retransmissions are wasteful
Rate ControlRate
Control
DestSource Congestion? Link loss?
Loss Rate
RTT
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 9
1. E2E Transport
E2E rate control is error-prone
E2E retransmissions are wasteful
Source DestRedundant
Transmissions
E2E Retransmissions
P
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 11
2. Packet as Unit of Control
Channel accessLink layer ARQ
Listen Backoff RTS/CTS
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 12
2. Packet as Unit of Control
Channel access Link layer ARQ
Timeout Backoff Timeout Backoff
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 13
3. Complex Cross-Layer Interaction
Transport
Link
Link-layer ARQ/backoff hurts TCP rate control
Highly Variable RTT
Link ARQ Link ARQ Link ARQ
Rate ControlRate
Control
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 14
Hop: Clean-Slate Transport Protocol
End-To-End
Packets
Complexity
Hop-by-Hop
Blocks
Minimize interaction
Block-switched networks: A new paradigm for wireless networks, Li et al. [NSDI 2009]
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 15
Hop Design
Reliable Block Transfer
ACK Withholding
Micro-block Prioritization
Virtual Retransmission
Backpressure
Per-hopMulti-hop
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 16
Reliable Per-Hop Block Transfer
B-SYN Request for B-ACK
B-ACK Packet bitmap
Mechanisms
Burst mode (TXOP)
Block ACK based ARQ
Benefits
Amortizes control overhead
CSMA TXOP
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 17
Hop Design
Reliable Block Transfer
ACK Withholding
Micro-block Prioritization
Virtual Retransmission
Backpressure
Per-hopMulti-hop
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 18
MechanismLeverages in-network cachingRe-transmits blocks only when unavailable in cache
BenefitsFewer transmissionsLow overheadSimple
Virtual Retransmission (VTX)
B-SYNB-ACK
A B
E
C
D
DATA
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 19
MechanismLeverages in-network cachingRe-transmits blocks only when unavailable in cache
BenefitsFewer transmissionsLow overheadSimple
Virtual Retransmission (VTX)
A B
E
D
B-SYNVTX
B-ACK
B-SYNVTX B-SYN with VTX flag set
VTX TimerE2E ACK
B-SYNVTX Timer
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 20
Hop Design
Reliable Block Transfer
ACK Withholding
Micro-block Prioritization
Virtual Retransmission
Backpressure
Per-hopMulti-hop
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 21
MechanismLimits outstanding blocks per-flow at forwarder
Backpressure
A B C D E
Source Dest
Slow
Limit on outstanding blocks=2
B-SYN
Hold B-ACK
B-SYNB-SYN
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 22
MechanismLimits outstanding blocks per-flow at forwarder
BenefitsImproves network utilization
Backpressure
A B C
D E
Source
Dest
F G
Dest
Aggregate goodput without backpressure: 6Mbps
20Mbps 10Mbps 20Mbps
20Mbps
20Mbps1Mbps
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 23
MechanismLimits #outstanding blocks per-flow at forwarder
BenefitsImproves network utilization
Backpressure
A B C
D E
Source
Dest
F G
Dest
Aggregate goodput with backpressure: 10Mbps
20Mbps 10Mbps 20Mbps
20Mbps
20Mbps1Mbps
Limit of Outstanding Blocks=1
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 24
Hop Design
Reliable Block Transfer
ACK Withholding
Micro-block Prioritization
Virtual Retransmission
Backpressure
Per-hopMulti-hop
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 25
Mechanism:
Receiver withholds all but one BACK
Benefit:Low overheadLess conservativeSimple
Ack Withholding
A C B
B-SYNB-ACK
DATA
B-SYN
B-ACKDATA
Withhold B-ACK
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 26
Hop Design
Reliable Block Transfer
ACK Withholding
Micro-block Prioritization
Virtual Retransmission
Backpressure
Per-hopMulti-hop
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 27
Micro-block Prioritization
MechanismsSender piggybacks small blocks to B-SYNReceiver prioritizes small block’s B-ACK
BenefitsLow delay for small blocks
SSH FTP
B-SYN
B-ACKDATA
B-SYN64B 1MB
Sender SenderReceiver
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 28
20 nodes on the 2nd floor of UMass CS building
Apple Mac Mini1.8GHz, 2GB RAM, Atheros 802.11 a/b/g card
Testbed
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 29
Single-flow Single-hop Performance
Hop achieves significant gains over TCP
HopTCP
28x
1.6x
1.2x
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 30
Single-flow Multi-hop Performance
HopTCP
2.7x
2.3x
1.9x
Hop achieves significant gains over TCP
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 31
Graceful Degradation with Loss
Emulated link layer losses at the receiver
TCP breaks down at moderate loss rates
HopTCP
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 32
Scalability to High Load
30 concurrent flows
Hop is much fairer than TCP
Meangoodput
HopTCP
Hop-by-hop TCP
150x
20x
2x
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 33
Hop Performance Breakdown
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 35
Low Delay for Small Transfers
AP
1 small transfer competing with 4 large
Small transfer size (KB)
Hop lowers delay across all file sizes
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 40
Related WorkFixing E2E rate-control
Separating loss/congestion [Snoop, WTCP, Westwood+, ATCP, TCP-ELFN][Snoop, WTCP, Westwood+, ATCP, TCP-ELFN]
Network-assisted rate control [ATP, NRED, IFRC, WCP][ATP, NRED, IFRC, WCP]
Hop circumvents rate control
BackpressureATM, theoretical work [Tassiulas et al.][Tassiulas et al.]
Tree/chain sensor data aggregation [Fusion, Flush][Fusion, Flush]
Reliable point-to-point transport [RAIN, CXCC, Horizon][RAIN, CXCC, Horizon]
Hop reduces backpressure overhead using blocksBatching
Common optimization at link [802.11e/802.11n, WiLDNet, Kim08, [802.11e/802.11n, WiLDNet, Kim08,
CMAP]CMAP], transport [Delayed-ACK, DTN2.5][Delayed-ACK, DTN2.5], and network [ExOR][ExOR] layersHop leverages batching across layers
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 41
Outline
Motivation
Block transport
Replication routing
Research challenges
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 42
Replication routing
PX Z
Y
P
P
W
P
Replication reduces delay under topology uncertainty
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 43
Routing as resource allocation problem
RAPID Protocol (X,Y):
1. Control channel: Exchange metadata
2. Direct Delivery: Deliver packets destined to each other
3. Replication: Replicate in decreasing order of marginal utility
4. Termination: Until all packets replicated or nodes out of range
YX
Change in utility
Packet size
Problem: Which packets to replicate given limited bandwidth to optimize a specified metric?
DTN routing as a resource allocation problem, Balasubramanian et al. [SIGCOMM 2007]
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 44
Objective: Minimize average delay
Define U(i) = -(T + D) T = time since created, D = expected remaining time to deliver
Simplistic assumptionsuniform exponential meeting with mean ¸global view
Utility computation example
D = ¸
iX
D = ¸/2
iY
D =¸/3
iW
Z
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 45
Utility computation example (cont’d)
Replicate i
Metric: Min average delay
Deadline of i < TDeadline of j = T1 > T
Metric: Maximize #packets delivered within deadline
Replicate j
iX
j
Y j
W Z
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 46
Deployment on DieselNet
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 47
Results based on DieselNet traces
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 48
Outline
Motivation
Block transport
Replication routing
Research challenges
UUNIVERSITY OF NIVERSITY OF MMASSACHUSETTSASSACHUSETTS A AMHERST • MHERST • Department of Computer Science Department of Computer Science 49
C1: When to replicate? Replication useful under
Topology uncertainty Delay optimization Light load scenarios
WLAN Mesh MANET DTN
Low uncertaintyForwarding suffices
High uncertaintyReplication useful
X1
X2
Xk
src dst
relays
Xi = r.v. for delay of path iForwarding delay = mini(E[Xi])Replication delay = E[mini(Xi)]
Claim: Delay benefit of replication is unbounded.