Dissertation Defense August 2002

Adaptive Networking Protocol for Rapidly Mobile Environments Ph.D. Dissertation Defense Edwin A. Hernandez August 2 th , 2002

Motivations Rapid Mobility is characteristic of forthcoming 3G networks Several micro-mobility and fast handoff protocols have been proposed Evaluation and design of such protocols are based on inadequate tool (ns-2) Ns oversimplifies important details to which rapid mobility is sensitive Each protocol has to be implemented into ns (slowing down innovation) Evolving wireless network technology (2.5G->3G) must be implemented into ns (also slow process) A Mobile Network Emulator is needed New Protocols can be created and tested without the need of a simulation environment

Network Simulator ( ns) The ns network simulator – Berkeley [Fall00] tcl/c++ object oriented, 20 Mbytes of code, wired and wireless network protocols

Table of Contents RAMON: The Emulation Alternative Performance of Mobile-IP in RAMON Predictable mobility – Ghost Mobile Node and Ghost Foreign Agents Summary of Findings Future Work

Publications E. Hernandez and A. Helal, " Examining Mobile-IP Performance in Rapidly Mobile Environments: The Case of a Commuter Train ," LCN 2001, Tampa, FL, Nov 14-16, 2001 E. Hernandez and A. Helal. “ RAMON: An Emulation Alternative ”, accepted LCN 2002 – workshop Wireless Networks. E. Hernandez and A. Helal “ RAMON: a network emulation testbed ”, submitted to Wireless Systems and Mobile Computing Journal, Wiley & Son’s, submitted August 2002. M. Chinta, A. Helal, E. Hernandez, “ ILC-TCP: Interlayer Collaboration Protocol ”, submitted MSWiM 2002.

Why Network Emulation? Pure simulation approach is inadequate [Paw, Jan02]&[Rappaport, May02] Simulation is more appropriate for higher layers, whereas emulation is more suitable for lower layers Emulation of emerging wireless network hardware is much faster and more accurate than oversimplification through simulation software Emulation of protocols is also faster and much more cost effective

RAMON: The Emulation Alternative Real wireless network hardware Software driven attenuators to emulate speed Emulation language to abstract descriptions of the scenarios to emulate A web service oriented GUI and tool for defining and controlling RAMON emulation scenarios

RAMON RAMON characteristics : Network emulation language to facilitate, academic and network-engineering work. Real implementation and code-extensions made to real mobility agents ns scripts can be parsed and emulated with minor modifications, a GUI can aid researchers in the definitions of mobility scenarios. Applications can be tested in rapid mobility conditions

Path loss attenuation with 802.11b access points in RAMON - n is the index of attenuation - It’s necessary two provide actual bandwidth to accurately estimate and reflect the effects of speed and handoff on network cards (a) Path loss and data rate for Cisco AP-350 (b) Path loss equations at different transmission power levels ( n =2.5)

Attenuation Control through Parallel Ports in RAMON

Emulation of speed Path Loss Equation: Scenario Attenuator 0 Attenuator 1 Attenuator 2 No connectivity -127 dB -127 dB -127dB One cell 0 dB <set < -80 dB -127 dB -127dB Two overlapped cells 0 dB < set < -80 dB 0 dB < set < - 80 dB -127 dB Three overlapped cells 0 dB < set < -80 dB 0 dB < set < - 80 dB 0 dB < set < - 80 dB

RAMON Emulation Language Updates attenuation and speed every X ms $granularity X N/A Sets the propagation model being used. $Propagation=”TwoRayGround”|”PathLoss”|any other. $set opt(prop) Propagation/TwoRayGround End of the emulation $end time $ns at $time end Starts after it’s called - $ns at $time start Sets the destination position and speed of mobile host. Acceleration = 0. $MH time x y speed $ns at $time [$MH etdest x y speed] Creates a Link between two interfaces using certain bandwidth and latency values $Link IP 1 IP 2 bw latency $ns duplex-link $node1 $node2 $bw $latency DropTail Creates a Wired Node with three interfaces. $WiredNode name IP 1 IP 2 IP 3 set wiredNode [$ns node $IP] The protocol being used $protocol=”MIP“ set mobile-ip 1 Sets the HA/FA at an IP address $HA name IP $FA name IP Set HA… /FA… The power level in mW in the access-point $BS name power=xxx set power 0.289 Sets an IP Address for the base-station $BS name IP= set BS [$ns node IP] Sets the coordinates of the Base-station $BS name X= $BS name Y= $BS X_ $BS Y_ Description Emulation script ns script

Sample Emulation Script $WiredNode node1 192.168.1.1 192.168.2.1 192.168.3.1 $WiredNode node2 192.168.2.2 192.168.4.1 192.168.5.1 $Link 192.168.2.2 192.168.2.1 10Mb 20ms $Link 192.168.1.1 128.227.127.11 10Mb 1ms … . $BS node7 X=250 Y=250 power=20dBm IP=192.168.7.1 $BS node8 X=750 Y=250 power=20dBm IP=192.168.8.1 $BS node9 X=1250 Y=250 power=20dBm IP=192.168.9.1 $BS node10 X=1750 Y=250 power=20dBm IP=192.168.10.1 $BS node11 X=2250 Y=250 power=20dBm IP=192.168.11.1 … $MH 0 1000 250 20m/s $start 10s $end-time 1500s $Propagation=”PathLoss” $Protocol “MIP”

Emulation Code Emulation( MH, granularity ); initializeResources( ); DetermineRoutes(route[][], time_end[], trajectory(MH)); while timer() > end_simulation do if timer>=timer_end[k] then k++; createRoute(route[k][1..3], time_end[k]); expireRoute(route[k-1][1..3]); emulateMovement(granularity, MH ); return;

NistNET Emulator for Wired Networks Wired network emulation required for academic and network engineering of rapidly mobile networks with may service providers and heterogeneous networks.

RAMON: The Prototype antennas Agents Access Points

Programmable Attenuators Controller for attenuator TTL-based/Parallel port Attenuators 0 – 128 dB

Graphical User Interface for RAMON Application in C# and .NET WinForms Easy to deploy as a web-service in the future. Remote experimentation can be available

GUI for RAMON Adding a Wired Node and a Link to the emulator

Table of Contents RAMON: The emulation alternative Performance of Mobile-IP in RAMON Predictable mobility – Ghost Mobile Node and Ghost Foreign Agents Summary of Findings Future Work

Experimentation Variables RAMON testbed Emulation scripts created with the tool and verified by hand. HUT (dynamics) Mobile-IP implementation, agent advertisement = 1 sec, hierarchical Mobile-IP, LFA and HFA required. Attenuators required of a special script to be turned off and eliminate the effect of the leaked signal. MAC handoff and roaming is within same domain. Handoff is not forced to the network card at the mobile node.

Experimental Results Signal strength measured with different values of n

Throughput and TCP Sequence Numbers Plot at 20 m/s (squared attenuation function)

Throughput and TCP Sequence Numbers ( n =2.5) at 20m/s TCP-sequence number-time plot is affected by the attenuation model

Throughput and TCP sequence Numbers ( n =3.5) at 20 m/s A sharper attenuation pattern results on a semi-linear shape.

Average Throughput at Different Speeds and Attenuation Patterns. In average the attenuation model selected will Affect the average throughput observed at different speeds. Throughput is a function of the signal strength received

TCP Sequence Numbers Plot at Different Speeds. (c) 20 m/s 40 m/s 80 m/s

Observations Handoff rate as equivalent of speed is not accurate, attenuation models are important The average value of throughput as speed decreased of at least 50% at 80 m/sec (compared to 20 m/sec) Other research [Camb01] showed a maximum of only 25% decrease at 20 handoff/min or an equivalent speed value of 300 m/sec. (assuming a cell diameter of 1000 m) RAMON can replicate realistic conditions of mobility

Predictive Mobility and Extensions to Mobile-IP Performance bottleneck of Mobile-IP: The mobile unit, requires of registration in order to maintain the home network aware of its mobility, the Home Location Register (HLR) and the home agent structure used in mobile-IP [Perk96a] Several experiments were conducted in RAMON and using an agent advertisement time of 1 sec, with no agent solicitation messages, handoff required approximately 2 to 10 seconds depending upon the speed of the mobile host, higher the speed higher the handoff.

Handoff in Mobile-IP Handoff overhead >= Registration Overhead Handoff Impact = confuses TCP

Reactive Mobile-IP (current implementation) Even the optimized, hierarchical M-IP (i.e. Lifix) relays on reactive mechanism and the agent advertisements to achieve mobility. Fast Handover is only equivalent to O(log( N) ) time on registration. Nothing is said about a more preemptive or predictive alternative for MIP GPS or Any Location Management Information is assumed to be available

The “ghost” Entities Ghost Mobile-Node (g-MN): As the mobile node moves along the different cells and follows a determined trajectory. A “virtual” repeater capable of registering and allocating resources in a predictive matter could potentially speed-up handoff and augment the performance of Mobile-IP and be able to cope with speed. The g-MN is cable of replicating the registration request, handling the creation of the tunnel, and replicating Authentication and Authorization information from the MN and act on behalf of the MN before is in the range of the new FA. Ghost Foreign Agent (g-FA): Similarly to the process and delegation of authority done with the MN, a g-FA could be created in the neighborhood of the FA. The g-FA will then advertise the FA presence of a different FA.

Ghost Mobile-IP Entities created as the MH moves They act on behalf of the MH and FA. Increase the “range” of the signal of the Mobile Host and Foreign Agent

Ghost Mobile-IP Anticipate movement and allocate resources before are required. One of the bottlenecks in the current M-IP implementation is the creation of the tunnel (ip-ip), takes a couple of seconds after registration is received by the HA or HFA. Kalman Filters can easily track and determine the position of a moving vehicle and be used to anticipate and preemptively allocate resources.

Ghost Mobile Node The g-MN creates a replica of the Registration UDP packet that the MN would have sent to the FA or to the HFA, depending on what it would have expected. Potential problems with nonce numbers and time-stamping (protection against reply-attacks). MD5 authentication and a shared key is required instead. Packet is created as a RAW socket, experimentally we observed the current registration packet and maintained the same values of “Lifetime”, Flags, and other Extensions used with Dynamics Mobile-IP implementation.

Ghost Mobile Node g-MN (Home Address, HomeAgentAddress) 1. w hile (true) do 2. FA ←FindClosestFA( MN ) 3.. if distance(FA, MN) > threshold then 4. HFA ←FindHighestFA(FA, HomeAgentAddress) 5. Register ( FA , Home Address , HFA ) 6. end 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type |S|B|D|M|G|V|rsv| Lifetime | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Home Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Home Agent | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Care-of Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + Identification + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Extensions ... +-+-+-+-+-+-+-+- distance(FA, MN) is the Predictive Distance , otherwise it is not predictive but reactive .

Ghost Foreign Agent Extends the range of the FA to allow other FAs in the vicinity to advertise on his behalf. ICMP Advertisement packets most be sent to the MH before it arrives to the FA so that the MH can add that FA to the list of potential FA to handoff. Very similar to the Shadow Cluster [Levi95] but in Mobile-IP not in Wireless ATM networks.

Ghost Foreign Agent g-FA ( ForeignAgen t) 1 . For each FA in Topology 2. i f (distance( FA , ForeignAgent )> threshold) and ( FA != ForeignAgent ) 3. SendGhostAgentAdv ( ForeignAgent, FA ) 4. end 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type | Length | Sequence Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Registration Lifetime |R|B|H|F|M|G|V| reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | zero or more Care-of Addresses | | ... | ICMP packet: Distance is always the same, predicted or current. FA speed = 0;

Location Tracking with Kalman Filter Kalman filters have been used in numerous applications ranging from location tracking and control of physical variables; wireless protocols are not the exception. D. Dailey, et. al.. [Dail00] solves the problem of tracking a vehicle and the time to arrival to a certain destination using the Kalman filter. The prediction done by the predictor is used to inform bus riders and anyone with a smart phone the waiting time of a bus route in Seattle, WA. The Kalman filter [Welc02] addresses the problem of trying to estimate the state: x R of a discrete-time controlled process that is governed by a linear stochastic difference equation.

Equations State vector : Measurement vector: In our case the state vector indicates, speed in <x,y> and the <x,y> Coordinates of the MH. The Measurement vector are the values of <x,y> measured from a GPS system or a location tracking device. System equations:

Equations : Matrices representing our system: Kalman Filter Time-Update equations: Q = E{ w k w k T }.

Equations Measurement-update equations Iteration between measurement-update and time-update Equations.

Filter Performance (MATLAB) Tracking Error and sampling rate X Y

Performance of M-IP and g-MN/FA The experiments corresponds to n =2.5 Average throughput much higher. Sampling at T =5sec Kbytes/sec Speed (m/s)

TCP sequence-time Plots at 40m/s Ghost-MobileIP Standard MIP Helsinki University of Technology – HUT Dynamics-0.8.1

Time-sequence Plots at 80 m/s Ghost MobileIP Standard MIP Helsinki University of Technology – HUT Dynamics-0.8.1

Observations For 40 m/s and n =2.5 we observed that TCP registered almost 20 million packets transferred, while in the non-predictive case about 14 million packets arrived from the FTP server. More than 10 million packets arrived to the mobile node using the predictive algorithm, while about 6 million made during the non-predictive case. This shows an improvement of approximately 1.5. Experiments were repeated at least 10 times for the average plot and the time-sequence plot represent the worst case observed.

Summary of Findings Mobile networking protocols, such as Mobile-IP, are not designed to handle high-speed gracefully. Mobility is reactive to the environment. Micro-mobility protocols, HAWAII and Cellular-IP, presented discrepancies in simulator. RAMON effectively replicates realistic conditions of showing results previously unknown and non-observed in simulation-based experiments

Summary of Findings RAMON : GUI + Emulation Language provide an abstraction for mobile emulators. Careful selection and capture of attenuation models showed different performance. Kalman Filters can track the location of a mobile host and this information used for handoff management.

Summary of Findings Extensions for mobile-IP called ghost-entities : ghost Mobile Node and ghost Foreign Agent improve the performance of MIP MIP was improved from a throughput of 60 Kbytes/sec to 90Kbytes/sec which represents almost 1.5 times increase with predictable protocol (at 80 m/s).

Future Work Stochastic techniques can be combined or compared with the Extended version of the Kalman Filter, as wells as other well known mechanisms such as Neural Networks and Machine learning. RAMON currently supports binary-tree scenarios only, skew or totally balanced topologies. RAMON must be extended to allow the emulation of any topology Signal leakage problems Buffering and packet forwarding to the g-MN.

Future Work Other protocols could benefit from the g-MN such as the Interlayer Layer Collaboration Protocol (ILC-TCP) [Chin02] which interacts with the lower layers of the stack to “freeze” TCP and acquired awareness of the wireless conditions Mobile IPv6 can be incorporated to the current stack and tested in the emulator Our current approach copes with speed but it doesn’t on the RTT of the packet which in our scenario increases per hop .

List of References [Camb00] A. T. Campbell, Gomez, J., Kim, S., Turanyi, Z., Wan, C-Y. and A, Valko "Design, Implementation and Evaluation of Cellular IP", IEEE Personal Communications, Special Issue on IP-based Mobile Telecommunications Networks, Vol. 7, No. 4, pp. 42-49, August 2000. [Cast98] C. Castelluccia. “A Hierarchical Mobile Ipv6 proposal”, Technical Report INRIA, France, November 1998 [Fall00] K. Fall, K. Varadhan, editors. NS notes and documentation. The VINT project, LBL, February 2000. http:// www.isi.edu/nsnam/ns / [Hern01] E. Hernandez and A. Helal, "Examining Mobile-IP Performance in Rapidly Mobile Environments: The Case of a Commuter Train," Accepted to LCN 2001 in Tampa, FL, Nov 14-16, 2001 [Levi95] D. Levine, I. Akyildiz, M, Naghshineh, “Shadow cluster concept for resource allocation and call admission in ATM-based wireless networks” Proceedings of the Annual International Conference on Mobile Computing and Networking, MOBICOM 1995. ACM, New York, NY, USA. p 142-150 [Perk95] C. E. Perkins, K. Luo “Using DHCP with computers that move”, Wireless Networks 1(1995) 341-353. [Perk96a] C. Perkins, IP mobility support, RFC 2002, IBM, October 1996 Perk96b] C. Perkins, Mobile-IP local registration with hierarchical foreign agents , Internet Draft, Internet Engineering Task Force (February 1996 [Ramj00] R. Ramjee, T. La Porta, S. Thuel, K. Varadhan, L Salgarelli, IP micro-mobility support using HAWAII , Internet draft submission , Jul 2000. [Solo98] J. Solomon. Mobile-IP. Prentice Hall, New Jersey, 1998

Dissertation Defense August 2002

More Related Content

What's hot (20)

Similar to Dissertation Defense August 2002 (20)

More from Dr. Edwin Hernandez (20)

Dissertation Defense August 2002

Editor's Notes