Block-switched Networks: A New Paradigm for Wireless Transport

Transcription

1 Block-switched Networks: A New Paradigm for Wireless Transport Ming Li, Devesh Agrawal, Deepak Ganesan, and Arun Venkataramani University of Massachusetts Amherst

2 What You Buy vs. What You Get TCP performs poorly over wireless links 2x 1.6x Advertised capacity: 11Mbps 40x

3 1. E2E Transport E2E rate control is error-prone E2E retransmissions are wasteful E2E route disruptions cause unavailability E2E Feedback Loss Rate RTT Rate Control Source Congestion? Link loss? Dest

4 1. E2E Transport E2E rate control is error-prone E2E retransmissions are wasteful E2E route disruptions cause unavailability E2E Retransmissions P Source Redundant Transmissions Dest

5 1. E2E Transport E2E rate control is error-prone E2E retransmissions are wasteful E2E route disruptions cause unavailability Route disruptions due to mobility

6 2. Packet as Unit of Control Channel access Link layer ARQ Listen Backoff RTS/CTS

7 2. Packet as Unit of Control Channel access Link layer ARQ Timeout Backoff Timeout Backoff

8 3. Complex Cross-Layer Interaction Link-layer ARQs/backoffs hurt TCP rate control Transport Rate Control Highly Variable RTT Link Link ARQ Link ARQ Link ARQ

9 Hop: A Clean Slate Re-design End-To-End Packets Complexity Hop-by-Hop Blocks Minimalism

10 Hop Design Multi-hop Per-hop Virtual Retransmission ACK Withholding Backpressure Micro-block Prioritization Reliable Block Transfer

11 Reliable Per-Hop Block Transfer Mechanisms Burst mode (TXOP) Block ACK based ARQ Benefits Amortizes control overhead B-SYN Request for B-ACK B-ACK Packet bitmap CSMA TXOP

12 Hop Design Multi-hop Per-hop Virtual Retransmission ACK Withholding Backpressure Micro-block Prioritization Reliable Block Transfer

13 Virtual Retransmission (VTX) Mechanism Leverages in-network caching Re-transmits blocks only when unavailable in cache Benefits Fewer transmissions Low overhead C Simple A B-SYN B-ACK DATA B D E

14 Virtual Retransmission (VTX) Mechanism Leverages in-network caching Re-transmits blocks only when unavailable in cache Benefits Fewer transmissions Low overhead Simple VTX Timer A B-SYN VTX B-ACK B B-SYN VTX E B-SYN with VTX flag set E2E ACK B-SYN D

15 Hop Design Multi-hop Per-hop Virtual Retransmission ACK Withholding Backpressure Micro-block Prioritization Reliable Block Transfer

16 Backpressure Mechanism Limits #outstanding_blocks per-flow at forwarder Slow A B C D E Source B-SYN B-SYN B-SYN Withhold B-ACK Dest Limit of Outstanding Blocks=2

17 Backpressure Mechanism Limits #outstanding blocks per-flow at forwarder Benefits Improves network utilization D A B C Source 20Mbps 10Mbps 20Mbps 20Mbps F 20Mbps 1Mbps E Dest G Dest Aggregate goodput without backpressure: 6Mbps

18 Backpressure Mechanism Limits #outstanding blocks per-flow at forwarder Benefits Improves network utilization D A B C Source 20Mbps 10Mbps 20Mbps 20Mbps F 20Mbps 1Mbps E Dest G Limit of Outstanding Blocks=1 Dest Aggregate goodput with backpressure: 10Mbps

19 Hop Design Multi-hop Per-hop Virtual Retransmission ACK Withholding Backpressure Micro-block Prioritization Reliable Block Transfer RTS/CTS is overly conservative and incurs high overhead.

20 Ack Withholding Mechanism: Receiver withholds all but one B-ACK Benefit: Low overhead Less conservative Simple A C B B-SYN B-ACK DATA B-SYN Withhold B-ACK B-ACK DATA

21 Hop Design Multi-hop Per-hop Virtual Retransmission ACK Withholding Backpressure Micro-block Prioritization Reliable Block Transfer

22 Micro-block Prioritization Mechanisms Sender piggybacks small blocks to B-SYN Receiver prioritizes small block s B-ACK Benefits Low delay for small blocks SSH 64B Sender Receiver Sender B-SYN B-SYN DATA B-ACK 1MB FTP

23 Testbed 20 nodes on the 2 nd floor of UMass CS building Apple Mac Mini Dual Core 1.8GHz, 2GB RAM, Atheros a/b/g card

24 Single-flow Single-hop Performance 1.2x Tcp 1.6x Hop 28x Hop achieves significant gains over TCP

25 Single-flow Multi-hop Performance 1.9x Tcp 2.3x Hop 2.7x Hop achieves significant gains over TCP

26 Graceful Degradation with Loss Emulated link layer losses at the receiver Tcp Hop TCP drops to zero with moderate losses

27 Scalability to High Load 30 concurrent flows 2x 20x Mean Goodput Tcp 150x Hop-by-hop TCP Hop Hop achieves massive gains over TCP and is much fairer

28 Hop over WLAN AP Mean (kbps) Hop improves utilization over TCP+RTS/CTS Median (kbps) Hop TCP TCP+RTS/CTS

29 Low Delay for Small Transfers 4 nodes perform large transfers, 1 node performs small transfer AP Transfer Size (KB) Hop lowers delay across all file sizes

30 Summary of Other Results Partitionable network TCP breaks down Hop significantly outperforms (TCP-based) DTN2.5 Network and link layer dynamics Hop outperforms TCP under dynamic network conditions Hop under g Similar performance gains as in b Impact on VoIP Hop impacts concurrent VoIP slightly more than TCP, but achieves significantly higher goodput.

31 Related Work Fixing E2E rate-control Separating loss/congestion [Snoop, WTCP, Westwood+, ATCP, TCP-ELFN] Network-assisted rate control [ATP, NRED, IFRC, WCP] Hop circumvents rate control Backpressure ATM, theoretical work [Tassiulas, ] Tree/chain sensor data aggregation [Fusion, Flush] Reliable point-to-point transport [RAIN, CXCC, Horizon] Hop reduces backpressure overhead using blocks Batching Common optimization at link [802.11e/802.11n, WiLDNet,, Kim08, CMAP], transport [Delayed-ACK, DTN2.5], and network [ExOR] layers Hop leverages batching across layers

32 Summary Block switching > packet switching Hop Key abstraction: Reliable per-hop block transfer Fast: Significant throughput, fairness, delay gains Robust: Degrades gracefully to challenged networks Simple: Minimizes complex cross-layer interaction Can we have one transport protocol for diverse wireless networks? Yes, we can! Source code at

33 The End Questions?

34 Dynamic Network Conditions 30 concurrent flows Auto Bit-Rate Control OLSR Hop continues to significantly outperform TCP under dynamic network conditions