Transcript ppt
A Comparison of Mechanisms for Improving Mobile IP Handoff Latency for End-to-End TCP MobiCom 2003 Robert Hsieh and Aruna Seneviratne School of Electrical Engineering and Telecommunications The University of New South Wales 26th February, 2004 Presented by Sookhyun, Yang Contents Introduction Related Works Experimental Methodology Experimental Results Conclusion 2/25 INTRODUCTION Mobility Related Terminology Mobile node (MN) Handoff (Handover) Layer 2 handoff Beacon message Access router (AR) Access network (AN) Mobile IP (MIP) – – – – – – Handoff latency Home network (HN) Foreign (Visited) network Home Agent (HA) Foreign agent (FA) Correspondent node (CN) Internet draft: http://www.ietf.org/internet-drafts/draft-ietf-seamoby-mobility-terminology-06.txt 3/25 INTRODUCTION Mobile IP (MIP) Foreign network (FN) reconfiguration IP’ COS FA CN (Care-of-address) When a MN moves and attach itself to another network – Need to obtain a new IP address – All existing IP connections to the MN need to be terminated and then reestablished IP Solution to this problem at MIP – Indirection provided with a set of network agents – Handoff latency tunneling binding intercept HA IP IP Home network (HN) Address reconfiguration procedure HA registration process – No modification to existing routers or end correspondent nodes Mobile node (MN) Access point (AP) 4/25 INTRODUCTION Motivation Effects of Mobile IP (MIP) handoff latency – Packet losses – Severe End-to-End TCP performance degradation Mitigation of these effects with MIPv6 extensions – Hierarchical registration management – Address pre-fetching – Local retransmission mechanism No comparative studies regarding the relative performance amongst MIPv6 extensions 5/25 INTRODUCTION Overview Evaluate the impact of layer-3 handoff latency on End-to-End TCP for various MIPv6 extensions – – – – – Hierarchical MIPv6 MIPv6 with Fast-handover Hierarchical MIPv6 with Fast-handover Simultaneous Bindings Seamless handoff architecture for MIP (S-MIP) Propose an evaluation model examining the effect of linear and ping-pong movement on handoff latency and TCP goodput Optimize S-MIP by further eliminating the possibility of packets out of order 6/25 RELATED WORKS Hierarchical Mobile IPv6 (HMIPv6) Minimize HA registration delay!! Internet RCOA_1 CN HA MAP AR AR MAP binding RCOA_2 binding AR AR AR AR AR AR AR AP AR RCOA_1 LCOA’ AP AP Micro mobility Access network RCOA_1 LCOA Macro mobility RCOA_2 LCOA’’ Access network Mobility Anchor Point (MAP) Internet draft - http://www.ietf.org/internet-drafts/draft-ietf-mipshop-hmipv6-01.txt 7/25 RELATED WORKS Local Handoff Latency Reduction Low latency address configuration – – – – Reduce address reconfiguration time Configure an address for MN in an network likely to move to before it moves Use L2 trigger Method Pre-registration – Perform L3 handoff before completion of L2 handoff Post-registration – – Setup a temporary bi-directional tunnel between oFA and nFA Allow MN to continue using oFA while registration at the time or later MIPv6 with Fast-Handover – – Combined method of pre-registration and post-registration Three phases 1. 2. 3. Handover initiation Tunnel establishment Packet forwarding 8/25 RELATED WORKS MIPv6 with Fast-Handover MN nFA oFA Beacon L2 trigger RtSolPr(Router solicitation proxy) PrRtAdv(Proxy router advertisement) HI(Handover initiation) F-BU(Fast-binding update) with COA Hack(Handover ack) 1 Handover initiation F-Back(Fast-binding ack) 2 Tunnel Establishment btw oFA & nFA F-BAck Disconnect Forward packets 3 Packet forwarding phase Connect F-NA(Fast neighbor advertisement) Deliver packets Internet draft - http://www.ietf.org/internet-drafts/draft-ietf-mipshop-fast-mipv6-01.txt 9/25 RELATED WORKS HMIPv6 with Fast-handover Combine HMIPv6 with Fast-handover Reduce latency due to address configuration and HA registration Relocate the forwarding anchor point from oAR to the MAP Internet CN HA MAP MAP AR Forwarding Forwarding AR AR nAR AR AR nAR oAR Access network Access network 10/25 RELATED WORKS Simultaneous Bindings Reduce packet losses N-casting packets with multiple bindings Forward packets for a short period to the MN’s current location and to n-other locations where the MN is expected move to Forwarding carried by oAR, MAP or HA nAR1 Simultaneous binding MAP oAR nAR2 AP (Access point) Internet draft- http://www.ietf.org/internet-drafts/draft-elmalki-mobileip-bicasting-v6-05.txt 11/25 RELATED WORKS Seamless Handoff for MIP (S-MIP) Provide a different approach to solve the timing ambiguity problem Build on HMIPv6 with Fast-Handover Use MN location and movement pattern to instruct MN when and how handoff is initiated Decision engine (DE) MAP – Store the history of MN locations – Determine movement pattern – Make “handoff decision” for MN DE nAR2 oAR nAR1 MN 12/25 RELATED WORKS Decision Engine MN location Tracking Linear <- Signal strength Handoff Decision Stochastic Stationary near the center Handoff mechanism 13/25 RELATED WORKS Handoff Mechanism Linear movement – Synchronized packet simulcasting (SPS) – Optimized S-MIP MAP optimization DE Stochastical manner – oAR and nAR are anticipation-mode – Maintain MN’s binding with oAR, nAR before F-NA – Reduce unnecessary re-setup S-packet F-packet oAR nAR Stationary state near the center – Establish multiple bindings with ARs – MN uses more than one COAs S-buffer F-buffer MN < SPS mechanism > 14/25 RELATED WORKS Optimized S-MIP Elimination of the possibility of packets out of order – Upon sending the F-BU to the oAR, MN must immediately switch to the nAR – After receiving F-BU, oAR must immediately forward packets to the nAR – oAR only needs to send the FBAck to the nAR IP packet filtering mechanism at nAR – – – – oAR incorrectly forwards IP packets with the S-bit set as f-packets Compare IP packets within the s-buffer and f-buffer at nAR Discard identical packets in s-buffer [optimized] Examine 16 bit identification, fragment offset, and flag fields in IP header 15/25 EXPERIMENTAL METHODOLOGY Implementation Simulator – Network Simulator version 2 (ns-allinone2.1b6a) Patch with the ns wireless extension module allowing basic MIPv4 Extension to the ns-2 – – – – – Mobile IPv6 protocol Hierarchical Mobile IPv6 protocol Fast-handover protocol Simultaneous bindings protocol Optimized S-MIP protocol Modification – Infrastructure mode: WaveLan with connection monitor (CMon) – Additional handoff algorithm: Midway handoff 16/25 EXPERIMENTAL METHODOLOGY Simulation Network Topology < Performance focus > • Handoff delay • TCP goodput • CN’s Congestion window Micro mobility Linear / ping-ping Overall handoff delay (D) = time(first-transmitted~retransmitted) +time(CN->MN) 17/25 EXPERIMENTAL RESULT – Handoff delay Sender (CN)’s view MIPv6 & HMIPv6 TCP sequence number • a: MIPv6 (resolution time 100ms) • b~e: HMIPv6 (resolution time 100ms) • f~I: HMIPv6 (resolution time 200ms) L2 handoff address resolution BU at MAP Out-of-sequence packet • MIP’s D = 814ms • HMIPv6’s D = 326ms Time (seconds) 18/25 EXPERIMENTAL RESULT – Handoff delay Fast-Handover Sender (CN)’s view TCP sequence number • f ~ i : fast-handover (resolution time 100ms) RtSolPr~PrRtAdv BU Proportional to distance (FA~HA) L2 handoff • D = 358ms • Even though forwarding mechanism, MN is unable to receive packets until the binding update is completed Time (seconds) 19/25 EXPERIMENTAL RESULT – Handoff delay HMIPv6 with Fast-Handover Receiver (MN)’s view TCP sequence number D = 270ms < CN’s cwnd > Packet forwarding Out-of-sequence packet Packet loss due to L2 handoff receive (data) send (ack) Time (seconds) 20/25 EXPERIMENTAL RESULT – Handoff delay S-MIP Sender (CN)’s view TCP sequence number <- Optimized S-MIP Hand off = 100ms No packet loss No out-of-sequence packet Time (seconds) TCP sequence number Non optimized S-MIP -> No packet loss Out-of-sequence packet Time (seconds) 21/25 EXPERIMENTAL RESULT Handoff Delay < Linear case > MIP 814ms HMIPv6 326ms MIPv6 with Fast-handover 358ms HMIPv6 with Fast-handover 270ms Simultaneous Bindings 268ms S-MIP (nonop) 0ms < Ping-pong case > • Completely break down • Affected to a lesser extent • Severe throttling • Excellent resilience S-MIP 0ms 22/25 EXPERIMENTAL RESULT TCP Goodput Linear : 1.447s PP: 14.23s MN is stationary near the PAR 23/25 EXPERIMENTAL RESULT Congestion Window Linear movement Ping-ping movement S-MIP Simultaneous Binding 24/25 Conclusion Analyze various handoff latency reduction framework Show the possibility of significantly reducing the latency by S-MIP Optimize the S-MIP scheme Future works – S-MIP under multiple connection scenarios – Scalability of the Decision Engine (DE) – Design more sophisticated positioning schemes for S-MIP 25/25