21
Managing TCP Connections in Dynamic Spectrum Access Based Wireless LANs Ashwini Kumar Prof. Kang G. Shin

Managing TCP Connections in Dynamic Spectrum Access Based Wireless LANs Ashwini Kumar Prof. Kang G. Shin

  • View
    219

  • Download
    1

Embed Size (px)

Citation preview

Managing TCP Connections in Dynamic Spectrum Access Based Wireless LANs

Ashwini Kumar Prof. Kang G. Shin

Overview

1. Introduction 2. Motivation & Challenges3. Main Contribution: DSASync

i. DSASync at Link Layer (DSASync_LL)ii. DSASync at TCP Layer (DSASync_TCP)

4. Evaluation5. Future Work

2/20

Dynamic Spectrum Access

• Communication increasingly becoming wireless and mobile, but– Wireless spectrum is a limited and shared resource– Side-effects: overcrowding, interference increase

• Dynamic Spectrum Access (DSA): Opportunistic wireless communication on licensed bands– Aims to solve spectrum-usage inefficiency &

shortage– Still evolving

3/20

4/20

Why DSA Can Be Effective?

[[Source: Shared Spectrum Company][[Source: New America Foundation]

Motivation

• WLANs mainly deployed as edge networks– Connections are from wireless client to cloud

• DSA is an inherently disruptive technology: blackout period (TFP)

5/20

B1

B2

T1 T2 T3 T4 T5 T6 T7 T8 Time

Raw

Cap

acity Sensing PU Access Channel-switchCapacity change

→ DSA contributes to capacity & delay fluctuations→ Device-centric QoS provisioning insufficient for end-to-end connections

Challenges

• Efficient monitoring/response mechanism to mitigate effect of DSA-related events

• Transparent to ongoing connections– End-to-end semantics of TCP must not be violated

• Ease of configuration/deployment– No changes must be introduced in wired cloud – No need to recompile/re-link of existing protocols

6/20

DSASync Overview

• No restrictions on the DSA protocol• Centralized operation at BS• Key concepts: caching + traffic-shaping

7/20

Wired Network (WN)

Corresponding Host (CH)

Base Station (BS)

Spectrum-agile Host (SH)

DSAN

DSASync

DSASync Architecture

• Logically a link layer component• But affects the transport layer, executes at BS• Key Advantage: TCP semantics maintained, no protocol modification

necessary - uses only existing TCP hooks

8/20

Application

Transport

Routing

Link/MAC

Physical

DSASync_TCP

DSASync_LL

DSASync Module

DSASync at Link Layer (DSASync_LL)

• Passive monitoring & estimation unit for DSA parameters – E.g., fsense(i), tsense(i), fsense(DSAN), tswitch(DSAN), gON(PU), etc– Uses MAC/PHY events or historical averages– Establishes if Transmission Freeze Period (TFP) is active

• Needed by DSASync_TCP component

• In implementation, usually encapsulated within the MAC

9/20

DSASync at TCP Layer (DSASync_TCP)

• Sniffs incoming/outgoing packets• Maintains per connection state, e.g., seq. nos. & last ACK

seen

10/20

CH-SH (Ingress) traffic manager

Capacity change manager

DSASync_TCP SH-CH (Egress) traffic manager

CH-SH (Ingress) Traffic Manager (DSASync_TCP_CH-SH)

11/20

Advt. 0 rwin to all CHs

Advt. 0 rwin for conn.

Start

Packet p received?

TFP going on?

Put p to transmit queue

Buffer p Buffer < thresh?

Conn. rwin filling up?

Yes

No

No

Yes

No

Yes

No

Yes

SH-CH (Egress) Traffic Manager DSASync_TCP_SH-CH

12/20

• Smoothens outgoing flow to mask DSA disruptions – Estimate outgoing data-rate D(i)– α(i) = 1-(TFPavg/ΔT), β(i) = max(α(i), αmin)

– Effective data rate, Deff(i) = β(i)D(i)

• Algorithm to empty queue at “smoothed rate”• Further optimization, from O(N) to O(1) FCFS

dequeue scheme

Capacity Change Manager (DSASync_TCP_CAP)

13/20

• Exploits built-in TCP congestion control– Executed when existing traffic suddenly

unsustainable – Force timeout or trigger fast retransmit

(recovery) by sending duplicate ACKs to CHs

Evaluation

14/20

• Implementation: Basic DSA protocol in 802.11 (MadWifi), PU emulation using MadWifi+Click– DSASync kernel module runs at the router

• Metrics: goodput, delay, jitter

• Testbed: infrastructure edge WLAN (6 clients)– Channel = 36, def. PHY capacity = 24Mbps, buffer capacity =

500MB– TCP Reno: initial TCP send & recv. window is 256KB– αmin = 0.5, Bhigh = 500MB, Blow = 400MB– Def. PU utilization = 20%, sensing overhead = 5%

Results: Overhead

15/20

• Overhead characterization:– Compare overhead with 802.11 (no DSA

overhead)– Avg. 1.9% reduction in throughput– End-to-end delay increase around 1.1ms

Microbenchmarks

16/20

• CH-SH (ingress) traffic benefitted most: 74% improvement, low retrans. overhead (0.018Mbps)

• SH-CH (egress) traffic improvement 10%

Microbenchmarks

17/20

• End-to-end delay variation lower for SH-CH (egress) traffic • DSASync makes SH-CH connection resilient

Microbenchmarks

18/20

• PHY capacity reduced by 50% at 5s

• Distributed DSASync agents useful?

Macrobenchmarks

19/20

• 4 TCP streams on each client – total 24 concurrent connections

• Trends similar to microbenchmarks

• Improvements even greater: 102% for CH-SH (ingress) direction

Future Work

20/20

• Evaluation on a large scale testbed

• Extend DSASync for UDP– Motivation: UDP streams carry most of the QoS-

sensitive multimedia traffic today – Challenge: stateless protocol implies no built-in

hooks to control the connection

Questions?