1. 1 MOBILITY AND SERVICE MANAGEMENT FOR FUTURE ALL-IP BASED WIRELESS NETWORKS Weiping He Preliminary Proposal, Dec. 12, 2006 Committee: Dr. Ing-Ray Chen, Committee Chair Dr. Csaba Egyhazy Dr. Mohamed Eltoweissy Dr. Chang-Tien Lu Dr. Gregory Kulczycki

2. 2 Outline Introduction Research Statement and Methods Related Works Dynamic Mobility Anchor Points IMSA: Integrated Mobility and Service Management Architecture Applications of Proxy for Integrated Cache Consistency and Mobility Management Conclusion and Future Work

3. 3 Mobility Management Enables networks to locate the MN for service delivery and to maintain active connections as the MN is moving. Location Management.  Keep track the location of MNs.  Include location registration and call delivery. Handoff Management.  An MN keeps the connection active when the MN moves.  Four tasks Deciding when to handoff Selecting a new AP Acquiring resources Informing the old AP reroute the packet and transfer state information.

4. 4 Service Management Ensures mobile nodes to get data services reliably, correctly and efficiently. Service request management  Request handling: accept service requests and transform requests into proper form.  Request delivery: forward server replies to the MN.  Request accounting, authentication and authorization (AAA) Service handoff management An MN keeps its services connection when it moves from one access point to another one.

5. 5 Research Statement Develop new mobility and service management schemes for future all-IP systems to minimize the overall network cost. Future all-IP based wireless networks provide network services based on the ubiquitous communication protocol: IP. Using per user based proxy to integrate mobility and service management to minimize the overall cost.

6. 6 Future All-IP based Wireless Network Architecture Home agent  Registration  Current location  Forward packets Correspond node Provides various services Access router  Offers IP connectivity to MNs  Powerful and flexible to host proxies to perform cross layers functions. Access Point Offers the wireless link connection to MNs

7. 7 Research Challenges and Motivations Mobile connectivity is highly variable. Mobile nodes are relative resource-poor. Workload to ARs is highly variable. Mobility and service characteristics of MNs are highly variable Vast majority of terminals will be mobile in a few years. The vast majority of traffic will originate from IP-based applications.

8. 8 Research Methods Extensive background research. Investigate of new techniques. Performance study via modeling and analysis. Demonstration of the applicability of proposed mobility and service management schemes. Simulation to validate analytical results.

9. 9 Contribution Propose and analyze per-user regional registration schemes for integrated mobility and service management. Given a set of parameters characterizing the operational and workload conditions of a MN, there exists an optimal regional area size for the MN such that the network communication cost is minimized. Our scheme outperforms basic Mobile IPv6, Mobile IPv6 Regional Registration, and Hierarchical Mobile IPv6.

10. 10 Network Layer Solutions (1) MIPv4 An MN is identified by its home address. If the MN is not in its home area, it has another address named Care of Address (CoA) associated with its current foreign location. The Home Agent maintains a dynamic mapping between the home address and CoA. A corresponding node always sends packets to the MN by the MN's home address. Pros: transparent to mobile applications. Cons: Triangle routing issue, CN  HA  FA  MN, slow handover.

11. 11 Network Layer Solutions (2) Mobile IP Regional Registration Purpose: to reduce the location handoff overhead. Moves within the regional registration area, MN only performs a regional registration to the GFA. Moves to another regional area, MN will perform a home registration. Packets path: CN  HA  GFA  FA  MN

12. 12 Network Layer Solutions (3) MIPv6 The MN determines its current location using the IPv6 router discovery protocol. The MN uses the IPv6 address auto configuration mechanism to acquire a care of address (CoA) on the foreign link. The MN notifies its home agent and CN for CoA change.

13. 13 Compare MIPv6 with MIPv4 Mobile IPv4Mobile IPv6 Mobile Node, Home Agentsame MN's home address Globally routeable home address and link-local home address Foreign Agent A “plain” IPv6 router on the foreign link (Foreign Agents no long exist) Foreign Agent care-of address vs. Collocated care-of address All care-of address are collocated Care-of address obtained via Agent Discovery, DHCP, or manually Care-of address obtained via stateless address autoconfiguration, DHCP, or manually Agent DiscoveryRouter Discovery Authenticated registration with home agent Authenticated notification of home agent and other correspondents Routing to mobile node via tunneling Routing to mobile node via tunneling and source routing Route optimization via separate protocol specification Integrated support for route optimization

14. 14 Hierarchical Mobile IPv6 (HMIPv6) AR AP CN AP AR AP AR Access network AR Internet HA Macro mobility Micro mobility MAP Mobility Anchor Point (MAP) RCOA_1 LCOA’ RCOA_1 RCOA_2 LCOA’’ RCOA_1 LCOA binding http://www.ietf.org/rfc/rfc4140.txt AR AP AR AP binding

15. 15 Application Layer Solutions Session Initiation Protocol ( SIP) An application layer protocol used to initiate, modify and terminate network sessions. Four elements: users agents, registrars, proxy servers and redirect servers. To support mobility.  SIP server in MN's home network receives registrations from the MN whenever the MN changes its location.  When the CN send an INVITE SIP message to the MN, the redirect server knows the current location information of the MN and forwards the INVITE message to the MN.  If a MN moves during an active session, it must send a new INVITE message to the CN using the same call ID. The new IP address is put in the contact field of the SIP messages. The CN will send future SIP messages to the new address.  If the MN is far away from the home network, every time it moves, it will send a new registration to the home SIP server. This may incur a high load.

16. 16 Summary of the mobility support approaches

17. 17 Service Management Approaches Result Delivery Protocol (RDP)  Using a service proxy to provide reliable message delivery to MNs. Created when a MN initiates a new series of service requests.  Provide a fixed location for the reception of server replies, keep track of pending requests, store the request results, and forward the results to the MSS.  Runs on the application layer, suitable only for connectionless request-reply communications.  The proxy moves whenever the MN moves across a location boundary, may incur a high communication cost.

18. 18 Service Management Approaches (continued) Mobile service management schemes based on location-aware mobile service proxies in PCS.  The personal proxies work as intelligent client-side agents to communicate with services.  The proxies cooperate with location management system, it is location-aware and can optimally decide when and how often it should move with the roaming user. Per-user integrated location and service management in PCS networks  A per-user service proxy is created to serve as a gateway between the mobile user and all client-server applications  The service proxy co-located with location database.  When there is a location handoff, a service handoff also happens to co-locate the service proxy with the location db. This allows the proxy to know the location of the mobile user to reduce the communication cost for service delivery.

19. 19 Service Management Approaches (continued) The above approaches are in the context of HLR/VLR based PCS networks,  MSS, VLR and HLR in PCS networks are powerful devices to perform both routing and computational functions. Routers in IP networks normally are specific routing devices.  PCS networks have regular shapes. IP subnets are shapeless.  Distance can be used to measure network cost in PCS. In IP networks, the network cost is normally measured by hops, which do not equal to distances.

20. 20 Dynamic Mobility Anchor Points Scheme Assumption: access routers are restricted to perform network layer functions. Determine best DMAP domain size per MN dynamically according to its mobility and service characteristics to reduce network and signaling cost Dynamic Mobility Anchor Points (Access routers chosen) for each MN MN determines dynamically when and where to launch DMAP for minimizing network cost DMAP domain size depends on MN’s mobility and service characteristics HA and CN know MN by RCoA

21. 21 DMAP (continued) After service area is crossed, MN selects AR of subnet just crossed as DMAP: MN determines size of new service area Obtains RCoA & CoA from current subnet registers (RCoA,CoA) to current DMAP by binding request message Inform HA and CN of new RCoA using standard Mipv6 Packet delivery route: CN->DMAP->MN (tunneling or direct)

22. 22 DMAP (continued) MN’s service area - K, IP subnets Goal : Dynamically determine optimal service area (K) per MN Special case :  K is constant for all MN’s Degenerates to HMIPv6  K is 1 Degenerates to MIPv6

23. 23 Diagram AR AP CN AP AR Access network AR Internet HA Macromobility Micromobility DMAP Dynamic Mobility Anchor Point (DMAP) RCOA_1 LCOA’ RCOA_1 RCOA_2 LCOA’’ RCOA_1 LCOA binding AP RCOA_2 tunneling

24. 24 Trade-off Large Service area : DMAP not change often  Communication cost for service data delivery high : CN->DMAP->MN  Location update cost is low Small Service area : DMAP changed often  Communication cost for service data delivery low  Cost of informing HA and CN of DMAP change is high

25. 25 Stochastic Petri Net Model

26. 26 C i,service : Network communication overhead to service a data packet when MN in i th subnet in service area Communication delay in the wireless link from the AR to the MN Delay between the DMAP and a CN in the fixed network Delay from DMAP to the AR of the MN’s current subnet in the fixed network Service Cost

27. 27 C i,location : Network signaling overhead to service a location handoff when MN in i th subnet in service area C location : Average communication cost to service a move operation by MN weighted by respective Pi probabilities i < K : MN inform DMAP of CoA change i = K : Location + Service to inform HA and N CNs of RCoA change Location Cost

28. 28 Total communication cost Total communication cost per time unit  : Data packet rate between MN and CNs   : MN ’ s mobility rate

29. 29 DMAP stays close to MN to avoid CN- DMAP-MN(service cost reduction) DMAP area large (mobility cost reduction) Degenerates to Basic MIPv6 DMAP degenerates to HMIPv6 Comparing DMAP with Basic MIPv6 and HMIPv6

30. 30 Cost difference curves are not sensitive to form of F(k) Assumption of F(k) justified Justify the assumption of F(K)

31. 31 IMSA: Integrated Mobility and Service Management Architecture Assumption: Access routers are powerful and flexible. Mobile proxies can be dynamically downloaded and roam the access routers to perform network layer and application layer functions on behalf of users and applications.

32. 32 IMSA on Mobile IP v4 A client-side proxy is created to serve as a GFA as in the MIP- RR to maintain the location information of the MN. The proxy will communicate with the correspondent node on behalf of the MN. The proxy will move only when the MN crosses a service area thus incurring a service handoff. The service area size depends on the mobility and service characteristics of the MN. Goal: network cost associated with mobility and service handoffs will be minimized.

33. 33 Message Flow in IMSA-MIPv4

34. 34 Service Handoff Process when Crossing a Service Area in IMSA-MIPv4

35. 35 Performance Model for IMSA-MIPv4

36. 36 Optimal Service Area Size Kopt with varying SMR and n ct in IMSA There exists an optimal proxy service area size to minimize the overall communication cost when given a set of parameter values characterizing the mobility and service behaviors of the MN and the network conditions of the Mobile IP network.

37. 37 Comparison of the IMSA-MIPv4 with Mobile IP v4. The total cost increases with the increase of SMR for both schemes Less communication overhead, especially pronounced when SMR is high.

38. 38 Comparison of the IMSA-MIPv4 with MIP- RR with Route Optimization. Total cost increases with the increase of either λor σ. IMSA-MIPv4 incurs less communication overhead than MIP- RR, especially pronounced when λor σis high.

39. 39 Optimal Service Area Size as a Function of n ct Initially increases. Context transfer cost becomes high, stay in a large service area to avoid handoff. The cost of context transfer would dominate the cost if n ct is large.

40. 40 Optimal Service Area Size as a Function of SMR When SMR increases, Kopt decreases. When SMR is small, the σis high compared to the λ; thus, the mobility management cost is larger than the service management cost. The proxy likes to stay at a larger service area to reduce the location handoff cost.

41. 41 IMSA on Mobile IP v6 IMSA-MIPv6 and the DMAP are similar. They differ only by the way of mapping a MN's RCoA to its CoA.  The DMAP design maps RCoA to CoA by having the current MAP maintain an internal table, so the MAP can intercept a packet destined for RCoA and forward it to the MN's CoA.  The IMSA-MIPv6 design maps RCoA to CoA by having a proxy run on the MAP directly receive a packet destined for RCoA, so the proxy can in turn forward the packet to the MN's CoA.

42. 42 PICMM: Proxy-Based Integrated Cache Consistency and Mobility Management Scheme in Mobile IP Systems Stateful cache consistency strategy: cache invalidation messages are asynchronously sent by the server the MN whenever data get updated. A per-user proxy to buffer invalidation messages to reduce uplink requests when reconnected. The proxy serves as a gateway foreign agent to keep track of the address of the MN in a region. Identify the optimal regional area size to minimize the overall network traffic cost, due to cache consistency management, mobility management, and query requests/replies.

43. 43 Cache invalidation strategies Stateful strategy:  When there is an update to a data object, the server will send an invalidation message to those MNs that keep a cache copy. Stateless strategy:  The server will broadcast information on data objects that have been updated either periodically or asynchronously. Problem: if an MN misses invalidation reports while it is disconnected, it will have to discard the cache content after it reconnects.

44. 44 The proxy’s three functions Working as a GFA as in regional registration to keep tracking MN's location; Acting as a service proxy for services engaged by the MN; Allocating a buffer space to store service context information for each MN. The proxy will receive invalidation reports from the server on behalf of the MN. If the MN is connected, the proxy will forward them to the MN. If the MN is disconnected, the proxy will store them in the buffer. Once the MN is reconnected, the MN will get the latest invalidation reports from the proxy.

45. 45 Integrated cache and mobility management scheme

46. 46 Cache invalidation process

47. 47 Query Process

48. 48 Disconnection Support

49. 49 Parameters The on/off (or wake/sleep) behavior of the MN: while the MN is in a wake state, it will go to sleep with rate ω w, while the MN is in a sleep state, it will wake up with rateω s. The residence time that the MN stays in a subnet while it is in a wake state. Service traffic between the MN and server applications

50. 50 Parameters

51. 51 Performance model

52. 52 Cost Function Derivation Total cost Effective data query rate is query arrival rate multiplied with the probability of the MN is being awake, λ Q is the aggregate query arrival rate Probability of the MN being in the awake state The effective mobility rate is the mobility rate multiplied with the probability of the MN being awake Effective data update rate is simply the aggregate data update rate

53. 53 Query Cost C query can be calculated as a weighted average in all states No query is issued while the MN is in sleep When the MN just wakes up, the MN will first check with the proxy for the cache status when answering a query, and, if cache miss, will get a copy from the server. If cache miss, the query cost will be MN  Proxy  CN Competition between effective query arrival rate and the update rate

54. 54 Mobility cost C mobility can be calculated as a weighted average in all states When the MN is in the sleep state, there is no location cost When the MN just wakes up, it will look for its proxy. If the MN does not move during sleep, the cost is contacting the current AR Just wakes up, If moves during sleep, proxy is moved to the current subnet, the cost includes context transfer cost and informing the HA and CNs the change of the proxy’s CoA If move within service area, MN only informs the proxy of the CoA change If moved K subnets, location handoff also triggers a service handoff, the cost includes context transfer cost and informing the HA and CNs the change of the proxy’s CoA

55. 55 Invalidation cost C invalidation can be calculated as a weighted average in all states If the MN is in sleep or just wake-up mode, then the invalidation report is buffered in the proxy. Otherwise, the cost is from the CN through the proxy to the MN

56. 56 Numerical results (1) Kopt vs. λ {q,i}. Kopt decreases slowly λ {q,i} increases. As λ {q,i} increases, the query cost increases, and subsequently the MN prefers a small service area size to reduce the query cost.

57. 57 Numerical results (2) Kopt vs. σ Kopt increases as σ increases. When the mobility rate is high, the mobility management cost is also high. The proxy likes to stay at a large service area to reduce the location handoff cost

58. 58 Numerical results (3) Kopt vs. the sleep ratio. Kopt decreases as the MN sleeps longer. Data in MN's local cache are more likely to be out- of-date when the MN sleeps longer. The MN will stay close to the proxy to reduce the triangular CN-proxy-MN communication cost

59. 59 Numerical results (4) Kopt vs. cache size. Kopt decreases as the number of cached data objects increases. As the cache size increases, more invalidation reports will be sent from the CN to the MN, given the same update rate for all data objects. To reduce the triangular CN-proxy-MN cost, the MN tends to stay closer to the proxy.

60. 60 Performance Comparison Compare our scheme with three schemes: No-proxy no-caching (NPNC) scheme: basic MIPv6 scheme Proxy no-caching (PNC) scheme: proxy-based regional registration scheme using a proxy for mobility management. No-proxy caching (NPC) management scheme: cached data objects maintained by the MN for cache management, no proxy used.

61. 61 Performance Comparison (2) No-proxy no-caching (NPNC) scheme proxy no-caching (PNC) scheme

62. 62 Generated Network Traffic as a Function of λ q,i Caching based schemes (NPC and PICMM) achieve much better performance, especially when λ q,i is large. NPC or PICMM increases slowly, PNC or NPNC increases drastically.

63. 63 Generated Network Traffic as a Function of σ PICMM scheme outperforms all other schemes, especially better when σ is high.

64. 64 Generated Network Traffic as a Function of Cache Size The total cost incurred under PICMM increases as the number of cached data items increases. Better performance especially when N data is large because caching saves much of the uplink cost for query processing.

65. 65 Generated Network Traffic as a Function of μ i The generated network traffic in non-caching based scheme (NPNC or PNC) is insensitive to μ i. The generated network traffic in caching schemes becomes sensitive to μ i. because the query traffic to the server depends on if cached data objects are valid. If data are updated frequently, a caching scheme will not perform good.

66. 66 Generated Network Traffic as a Function of ω s /ω w. When the sleep ratio is ω s /ω w extremely large, MN is mostly sleeping all the time. The cache is mostly invalid due to long sleep. PICMM would perform worse than PNC because of the CN- proxy-MN triangular cost for routing query inquiries/replies and invalidation reports under PICMM. At a reasonable range of sleep ratio (<2), PICMM outperforms all other schemes.

67. 67 Publications Conferences Paper  I.R. Chen, W. He, and B. Gu, DMAP: A Scalable and Efficient Integrated Mobility and Service Management Scheme for Mobile IPv6 Systems, 2nd IEEE International Workshop on Performance Management of Wireless and Mobile Networks, Tempa, FL, November 2006 Submitted papers  I.R. Chen, W. He, B. Gu. Proxy-based Regional Registration for Integrated Mobility and Service Management in Mobile IP Systems. (Submitted to Computer Journal)  I.R. Chen, W. He, B. Gu. IMSA-MIPv6: Integrated Mobility and Service Management Architecture for Mobile IPv6 Systems. (Submitted to Wireless Personal Communications Journal)  W. He, I.R. Chen, B. Gu. A Proxy-Based Integrated Cache Consistency and Mobility Management Scheme for Mobile IP Systems (Submitted to the IEEE 21st International Conference on Advanced Information Networking and Applications )

68. 68 To be Completed and Future Work Evaluate our designs by simulation with ns-2. Extend two-level regional registration design to more than two levels as in HMIPv6 and identify the optimal level to use to maximize the system performance. Investigate context-aware database applications that can benefit from knowledge of the MN's location and service context. Add fault tolerance and recovery into our designs. Experiment with mobile database query applications and the design of PICMM to decide which it is more beneficial whether to forward a copy of the data object instead of an invalidation report to the MN.

69. 69 Schedule DeadlineResearch Activity January, 2005Surveyed and analyzed existing mobility/service management in wireless networks. May, 2005Studied the characteristic of future all-IP based wireless networks. September, 2005Proposed and analyzed IMSA-MIPv4: proxy-based integrated mobility and service management scheme in Mobile IPv4 systems January, 2006Proposed and analyzed IMSA-MIPv6: proxy-based integrated mobility and service management scheme in Mobile IPv6 systems April, 2006Proposed and analyzed DMAP for the case in which ARs can perform only network-layer functions: DMAP November, 2006Proposed and analyzed PICMM for the case in which ARs can perform application-layer functions allowing proxies to carry service context information regarding cached data objects for mobile database applications. December, 2006Preliminary exam

70. 70 Schedule (continued) March, 2007Completion of design and analysis of hierarchical DMAP and IMSA for MIPv6 June 2007Completion of design and analysis of fault tolerance and recovery of DMAP and IMSA for MIPv6 September, 2007 Completion of extension to PICMM to deal with the case that data objects are forwarded to the MN instead of invalidation reports for mobile database query applications November, 2007 Completion of the applicability study by identifying context aware database applications that can benefit from DMAP and IMSA designs; completion of algorithm design and analysis for such applications identified. December, 2007Research Defense March, 2008 Completion of simulation studies based on ns-2 to validate analytical results as well as to compare DMAP, IMSA and PICMM and new algorithms extended against basic MIPv6, MIP-RR, HMIPv6 and other existing algorithms in MIPv6 for mobility and service management. May, 2008Final defense.

71. 71 Thank you