1

Disruption-Tolerant Link-level Mechanisms for Extreme Wireless Network Environments

Vijay Subramanian1, K. K. Ramakrishnan2 and Shiv Kalyanaraman1

1-(Rensselaer Polytechnic Institute) , 2-(AT&T Labs Research)

Status

Reports

Packets,FEC

Repairs

TCP Sender TCP Receiver

We gratefully acknowledge support from AFOSR ESC Hanscom and

MIT Lincoln Laboratory, Letter No. 14-S-06-0206

2

Growth of Wireless NLOS Mesh Deployments

Taipei

Philadelphia

San Francisco

3

Wireless Mesh Disruptions: Time-Scales & Reslience Strategies

1-10s

Bit errors

10ms-1s

Interference, Capture Effects, Jamming(Short-Medium term)

1-100sec

Longer-term path/link disruptions

Packet Erasures, Burst losses

10 s-1ms

Basic FEC Adaptive Hybrid ARQ/FEC(link/transport)LT-TCP, LL-HARQ

Cross-layer functions(transport/link/MAC)

Cross-layer functions+ multi-path transport(transport/routing/link/MAC)

In this paper, we look at link- and cross-layer protocol design under medium time-scale disruption conditions (10ms-1s).

Wireless links of NLOS meshed networks susceptible to poor performance &

outage due to path loss, shadowing, fast and multi-path fading, and interference

4

Recent Work: LT-TCP (Small Time-Scale)

• SACK: Exponential falloff in performance with PER

• 5%+ PER • 100 ms+ RTT

• LT-TCP: Linear falloff in performance with PER

• Scales well for even up to 50% PER • Scales well with RTT (100ms)

5

LT-TCP, LL-HARQ Scheme Features

• Initial transmission consists of data + PFEC packets.

• Feedback from the receiver indicates the number of units still needed for recovery.

• RFEC packets are sent in response to the feedback.

• If k out of n units reach the receiver, the data packets can be recovered.

• LT-TCP at the transport layer and LL-HARQ at the link layer.

• LL-HARQ operates with a strict limit of 1 ARQ attempt to bound latency.

6

Medium Time-Scale: Issues

• Cross-system Interference and Co-channel interference are the main causes of packet loss.– Residual loss rate on these wireless links can be non-

trivial.– Over multiple hops, the end-end loss rate can be

significant.

• End-End residual loss leads to timeouts at TCP-SACK, low throughput/goodput and poor utilization on an end-end basis.

7

Medium Time-Scale: Cross-Layer Issues

• PHY/MAC protocols can confuse interference with noise and perform rate-adaptation (a.k.a. adaptive modulation/coding). – With rate-adaptation, packets can be on the air longer. – In CSMA-based MAC, the same diversity modes (time-, frequency-, code-)

are being competed for by all users to achieve high bit-rate– The “collision” periods will now be longer and reducing goodput & increasing

latency.– We experiment with turning off rate adaptation (in practice, when interference

is sensed)

• MAC backoff also increases ARQ delays at the link layer• Variable packet transmission time => large and variable queuing delays,

confusing the AQM scheme.• Disruptions more than several RTTs => LL-HARQ cannot overcome &

we propose a mode-switching algorithm to complement it.

8

Disruptions Cause:Interference Capture Effects in 802.11 Links

9

Link-Layer Interactions and Interference

• 802.11b WLANs operate in unlicensed ISM band– 2.412 - 2.480 GHz– Can operate at 11 , 5.5 , 2 or 1 Mb/s raw data rate with rate

adaptation algorithms that are typically proprietary.

• Smaller Cells without fine-grained adaptive power-control: Higher probability of interference.– Rate adaptation counter-productive with CSMA/CA MAC.

• Cross-System Interference– Other systems that use the unlicensed ISM band include

Bluetooth , cordless phones, microwaves etc..

• Co-Channel Interference– Other WLAN systems using the same frequency in close

proximity.– Hidden Nodes in Remote Cells

10

Co-Channel Interference: Simulation Setup

• Node 1 is uploading data through BS-1• Node 2 is downloading a large file from BS-2

– Capturing the channel

• Node 1 is effectively experiencing capture for 250 ms every 2 seconds.

11

Results for Co-Channel Interference

• When RTT > 100 ms, the residual loss of even one WLAN hop

(subject to capture effects) can lead to low TCP-SACK throughput:

• LT-TCP + modest parameter changes at the link/MAC restores performance.

12

Disruptions (100% loss) With High Loss (0-50%) when no disruptions

Possible Causes: Outage due to path loss/shadowing/interference (cell edges), mobility

13

Simulation Setup: 1-hop and 4 hops

ON Period: 100% Loss (Disruption)

OFF Period: 0-50% Loss (High Loss!)

14

Performance of LL-HARQ with Longer Disruptions

To mitigate disruptions, link enhancements are needed even with LL-HARQ at the link-layer.

Disruption ON/OFF Model: 100ms in ON State (100% loss).

100ms in OFF state (p% loss).

LL-HARQ at links. Average PER shown in x-axis.

15

LL-Mode-Switching with Disruption Detection

The link operates in either pipelined mode (no outage)or in stop-wait “probing” mode (outage detected). “Outage” = all fragments of a packets lost. [ HARQ limit of 1 applies to pipelined mode only. Separate ARQ limit for stop-wait probing mode.]

16

TCP-SACK and LT-TCP Performance under Disruptions with LL-Mode-Switching

Disruption-tolerance (Mode-switching) enhancements to LL-HARQ: Low per-hop residual loss rate (even for 50%-100% ON-OFF case)!LT-TCP still useful for multiple hops to deal with accumulated loss rate

17

Trade-off between ARQ and Performance (with Disruptions)

During disruptions, having unbounded ARQ attempts (w/ stop-and-wait mode) is counter-productive due to spurious timeouts (despite having lower residual loss) !

18

TCP-SACK and LT-TCP Performance under Disruptions with LL-Disruption Enhancements (I)

19

Link and Transport Throughput and Goodput over 1 hop and 4 hp scenarios.

20

Summary

• LT-TCP provides robustness even under conditions of large and bursty loss rates.– Avoids timeouts– High Goodput– Increased Dynamic Range

• TCP performance over wireless with residual erasure rates 0-50% (short- or long-term).

• Outage and Disruptions at link-level impact the performance of transport layer protocols.

• Outage detection and protection at link layer can improve performance even under severe conditions (with support at the transport layer).

21

Thanks!

Researchers:

•Vijay Subramanian: •subrav@rpi.edu (Rensselaer Polytechnic Institute)

•K.K. Ramakrishnan, •kkrama@research.att.com (AT&T Labs Research)

•Shivkumar Kalyanaraman: •shivkuma@ecse.rpi.edu (Rensselaer Polytechnic Institute)

Thanks also for the support from AFOSR ESC Hanscom and MIT Lincoln Laboratory, Letter No. 14-S-06-0206

22

Extra Slides

23

Basic Link-level Scheme in more Detail

24

Link-level Insights

• Link Level Recommendations include:– Moderating the rate adaptation technique as a

function of interference detection– Larger buffers with flexible AQM/ ECN markings– Enhance TCP with LT-TCP mechanisms to be

robust in high loss rate scenarios.– Link-level can provide persistence beyond

disruption time-scales

Link-level ARQ is not a panacea even with LANs since end-end latency (spurious timeouts), and residual loss matter.

But LL can provide persistence across a short-term disruption period using mode-switching.

25

Future Research Directions

• Test and validate our approaches using real-world traces and data sets.

• Compare our proposed approach with other existing schemes such as SCTCP, WTCP etc..

• Move towards a real-world implementation to study the impact of practical constraints.– Proposed platform is Linux (2.6+)– This will help quantify processing costs (memory

and time) and ensure backward compatibility.

26

Co-Channel Interference

• Future cells could be small and compact to provide high data rate to users.

• Users may be able to connect to multiple base stations (some using the same frequency)

• RTS/CTS may also not be able to help if interfering node is far.

• Assume rate-adaptation is turned off and cells operate at 11Mb/s with only the MAC transmission rate at 1 Mb/s.

How can end-end mechanisms, such as TCP enhancements

help to make

communication robust under these high loss scenarios?

27

Interference & Capture: Negative Effects

• Co-channel interference (many users, unplanned environments)

• Cross-system interference (eg: Wifi vs Bluetooth, microwave ovens, jamming)

• MAC protocols like CSMA/CA => some users can consistently “capture” the channel– Unfairness, temporary outage (100s of ms)

• Link-level disruptions can also be caused due to mobility, temporarily misaligned directional antennas etc

Rate-adaptation with CSMA/CA and high load: harmful if the source of packet loss

is interference rather than fading => we turn off rate-adaptation if interference.