1 Performance Analysis of a Preemptive and Priority Reservation Handoff Scheme for Integrated Service- Based Wireless Mobile Networks by Jingao Wang, Quing-An Zeng, and Dharma P. Agrawal Presented by Okan Yilmaz CS 6204 Mobile Computing Virginia Tech Fall 2005

2 Abstract  Analytical Model & Performance Analysis  Call Types: Originating calls Handoff requests  Service Types: Real-time Non-real-time  Partitioning based system model Real-time service calls only Non-real-time service calls only Handoff requests only  Preemptive priority handoff scheme

3 Abstract (cont)  Multidimensional Markov Model to estimate Blocking probability of originating calls Forced termination probability of handoff calls Average transmission delays  Simulation and Performance Analysis Different call holding times Several cell dwell time distributions  Results Significantly reduces the forced termination probability of real-time calls Negligible packet loss of non-real-time calls

4 Introduction  2G Networks Limited and far from acceptable  Voice  Short message  Low speed data  3G Networks Demand for Integrated services  Business customers Any time, any place Employees, key customers e.g., brokerage, banking, emergency services, traffic reporting, navigation, gambling, etc. Wireless and VLSI Technology  Multi-media-ready cell phones, pocket PCs, Palms

5 Challenges of Integrated Services  True combination of real-time and non-real-time services  Maximize the utilization of network infrastructure  Quality of service (QoS)  Handoff handling Forced termination of an outgoing call is more annoying than blocking of a new call

6 Handoffs  Handoff: changing parameters of a channel Frequency, time slot, spreading code, or combination of them  When: crossing cell boundary or deteriorating signal quality  Cell structure Support a drastic increase of demand  Microcell, picocell, hybrid cell  Smaller cells  More handoffs

7 Handoff Design Issues  Forced termination versus new call blocking  Increased channel utilization in a fair manner  Goal: Minimization of forced termination of real-time service Without drastically sacrificing the other QoS parameters  Several studies based on voice based cellular networks  Need for support of multiple service types simultaneously  Keys for a good scheme: Service dependent  Delay sensitivity: non-real-time versus real-time  Preemptive model: priority reservation handoff

8 SYSTEM MODEL  Homogenous cell with fixed number of S channels  Reference cell approach  Call types: Real-time originating call: MU dials a number to place a real-time call Real-time handoff request: MU holding a channel enters the handoff area Non-real-time originating call: MU places a non-real- time call Non-real-time handoff request: Non-real-time MU holding a channel approaches and crosses a cell boundary  Cell boundary: The points where the received signal strength between two adjacent cells is equal

9 Notation  OR : arrival rate of real-time originating calls  HR : arrival rate of real-time handoff requests  ON : arrival rate of non-real-time originating calls  HN : arrival rate of non-real-time handoff requests  RC: real-time service channels group with capacity S R  CC: common handoff channels group with capacity S C  NC: non-real-time service channels group with capacity S N  RT only: In CC, real-time service channels reserved exclusively for real-time handoff calls with capacity S E  CH: In CC, channels that can be used by both real-time and non-real-time handoff calls with capacity S C - S E  RHRQ: real-time service handoff request queue with capacity M R  NHRQ: non-real-time service handoff request queue with capacity M N

10 System model for a reference cell  OR RC(S R )  HR RC(S R )  HC(S c -S c )  RT(S E )  RHRQ(M R )  HN NC(S N )  HC(S c -S c )  NHRQ(M N )  ON NC(S N )

11 Algorithm for Originating Calls

12 Algorithm for Handoff Requests

13 System Design (cont)  Preemptive procedure: real-time handoff request calls preempt non-real-time handoff request calls if a non- real-time in CC and NHRQ is not full  Real-time handoff requests may preempt non-real- time handoff requests irrespective of NHRQ being full or not No need if very large NHRQ buffer  Real-time handoff request are dropped If RHRQ is full (both RHRQ and NHRQ are full in preemptive scheme) If the handoff request in RHRQ cannot get service until it moves out of the handoff area

14 System Design (cont)  Non-real-time handoff requests will never be dropped If NHRQ is large enough (not necessarily be infinite)  Because the non-real-time handoff request is transferred from the reference cell to another cell  Waiting time in NHRQ = dwell time of non-real-time service subscribers  Real-time handoff request calls can continue until signal strength becomes not enough to get service This is ignored in paper. It is assumed that the call is blocked.

15 Traffic Model  Three characteristics: Call arrival process Call holding time Cell dwell time  Call arrival: Poisson process  Call holding time and cell dwell time Two approaches:  Traffic model: general independent identically distributed (i.i.d.) Exponential, gamma, lognormal, hyper-exponential, hyper-Erlang  Analytical model: User’s mobility, the shape and size of the cell, and exponential distribution are used to determine cell dwell and call holding time  Paper uses the second for analytical modeling, both for numerical and simulation results

16 Dwell Time  Two-dimensional fluid model f V (v): pdf of the speed V of MU E[V]: mean of the speed of MU  MU moves randomly any direction in [0,2)  Assumes uniform density of users

17 Cell Dwell Time  : density of MUs in the cell  N O : number of cell outgoing MUs with moving speed v and v+v  N T : total number of cell outgoing MUs per unit time  A: area of the cell  L: length of the perimeter   dwell : average outgoing rate of an MU within a cell  T dwell : cell dwell time with a random exponential distribution with mean 1/ dwell  Biased sampling theory in boundaries [1]

18 Handoff Area Dwell Time  f V* (v): pdf of the speed of real-time service subscribers crossing cell boundary V*  D: the length of moving path of mobile users in the handoff area  T h : dwell time of real- time service subscribers in the handoff area  E[T h ]: Average handoff area dwell time  Path length and velocity of MUs are independent

19 Channel Holding Time  Exponential distribution T CR : Call holding time of real-time calls T CN : Call holding time of non-real-time calls  CR : Service rate of real- time calls  CN : Service rate of non- real-time calls T R : Channel holding time of real-time service calls T N : Channel holding time of non-real-time service calls

20 Arrival Process of Service Calls  Poisson process  OR : arrival rate of real-time originating calls  HR : arrival rate of real-time handoff requests  ON : arrival rate of non-real-time originating calls  HN : arrival rate of non-real-time handoff requests  Need to compute HR and HN from OR and ON, respectively  Homogenous mobility pattern Mean number of incoming handoffs to reference cell = mean number of outgoing calls from the reference cell

21 Arrival Process of Service Calls (cont)  E[C R ]: average number of real-time calls holding channels in the reference cell  OUTR : departure rate of real-time handoff calls from the reference cell

22 Arrival Process of Service Calls (cont)  E[N N ]: average number of both non-real-time service requests and calls in the reference cell E[C N ]: average number of non-real-time MUs holding channels in the reference cell E[L N ]: average length of NHRQ : total arrival rate of calls

23 M/M/3/3  M/M/3/3 [2]: M: Exponential or Poisson arrivals M: Exponential or Poisson service 3: Number of servers 3: Maximum number of customers in the system  P 0 + P 1 + P 2 + P 3 =1  (+) P 1 = P 0 + 2 P 2  P blocking = P 3  Throughput = (1-P 3 ) *

24 PERFORMANCE ANALYSIS i j k l m

25 Stable State diagram for (i=1, j=1, k=1, l=2, m=0) S = S R + S C + S N =12 S R = 6; S C =S N =3; S E =1 M R =5; M N =50; N T =3162

26 Total number of states  Four cases to consider: 1.Both RHRQ and NHRQ are empty:  0≤ i ≤S R ;0 ≤ j ≤ S c - k; 0≤ k ≤ S c - S E ; 0≤ l ≤ S N ; m = 0 k=0  j=(0.. S c ) : S c +1 possibilities k=1  j=(0.. S c -1) : S c possibilities … k= S c -S E  j=(0.. S E ) : S E +1 possibilities Total = [(S c -S E +1) * (S c + S E +2)]/2 states  N 1 =[(S R +1)*(S c -S E +1)*(S c + S E +2)*(S N +1)]/2 states 2.RHRQ is not empty while NHRQ is empty:  i = S R ; S c < j + k  i =S R ; S c -k+1≤ j ≤ S c + M R + k; 0≤ k ≤ S c -S E ; 0≤l ≤S N ; m=0 k=0  j=(S c S c + M R ) : M R possibilities k=1  j=(S c.. S c + M R + 1) : M R possibilities … k= S c -S E  j=(S E S c + M R ) : M R possibilities Total = [(S c - S E +1) * M R ] states  N 2 =[(S c - S E +1) * M R * (S N + 1)]/2 states

27 Total number of states (cont) 3.RHRQ is empty NHRQ is not empty:  S c -S E ≤ j + k; l = S N ;  0≤i ≤S R ; S c -S E -k ≤ j ≤S c -k; 0≤k ≤S c -S E ; l=S N ; 1≤m ≤M N k = 0  j=(S c - S E.. S c ) : (S E +1) possibilities k = 1  j=(S c – S E S c - 1) : (S E +1) possibilities … k = S c - S E  j=(0.. S E ) : (S E +1) possibilities Total = (S c - S E +1) * (S E +1) states  N 3 = (S R + 1) * (S c - S E + 1) * (S E + 1) * M N 4.Both RHRQ and NHRQ are not empty  i = S R ; S c < j + k; l = S N ;  i = S R ; S c -k+1 ≤ j ≤S c + M R - k; 0≤k ≤S c -S E ; l=S N ; 1≤m ≤M N k = 0  j=(S c +1.. S c + M R ) : M R possibilities k = 1  j=(S c.. S c + M R - 1) : M R possibilities … k = S c -S E  j=(S E +1.. S E + M R ) : M R possibilities Total = [(S c -S E +1) * M R ]/2 states  N 4 = [(S c -S E +1) * M R * M N ]/2 states

28 Normalizing Condition 1.Both RHRQ and NHRQ are empty 2.RHRQ is not empty while NHRQ is empty 3.RHRQ is empty while NHRQ is not empty 4.Both RHRQ and NHRQ are not empty

29 Average number of calls  E[C R ]: average number of real-time calls holding channels in the reference cell  1&3: i + j: real-time calls  2&4: RC is full; S C -k real- time calls  E[N N ]: average number of both non-real-time service requests and calls in the reference cell  1&2: k + l: non-real-time calls  3&4: RN is full; S N +k real-time calls; m calls in NHRQ

30 Pseudo-code to solve (N T +2) independent nonlinear equations

31 Blocking Probabilities  Originating real-time calls are blocked when i= S R  Forced termination of real-time service handoff requests B HR : Blocking probability  M R is full D R : dropping probability  M R is not empty

32 Channel and RHRQ buffer utilizations  Utilization=mean channel used/ S  E[ C N ]: average number of calls holding channels 1&2: k+l: non-real-time calls 3&4: NC is full; S N +k real- time calls; m calls in NHRQ  RHRQ utilization = mean number of channels in RHRQ/M R  E[L R ]: average length of RHRQ 1&2: j+k-S C real-time handoff requests waiting in RHRQ

33 NHRQ Buffer Utilization and Forced Termination probability  NHRQ utilization = mean number of channels in LHRQ/M N  E[L N ]: average length of NHRQ 1&2: m non-real-time handoff requests waiting in NHRQ  P h : Probability that a real- time service call triggers a handoff request in the reference cell Real-time service call holding time > the cell dwell time  P hf : Forced termination probability of real-time handoff calls (l-1) successful handoff followed by a forced termination

34 Transmission Delay of non-real- time service  T d : The lifetime transmission delay of non-real-time service Sum of T w s  T w : transmission delay on non- real-time service in each cell  Little’s Law Mean waiting time = mean number of customers in queue / throughput  B ON : blocking probability of originating non-real-time calls 1 - P[NCS N ]  E[T S ]: Average serving time of non-real-time calls (mean number of calls getting service + in queue) / (total throughput)  B HN : blocking probability of non- real-time service handoff requests NHRQ is full: m = MN

35 Average transmission delay of non- real-time service (cont)  N h : average number of handoff per a non-real- time handoff request (delay due to N h handoffs + call holding time) by average serving time  E[T N ]: average transmission delay of non-real-time service Handoff arrival probability times average delay each handoff request ecounters

36 Numerical and Simulation Results  Integrated service homogenous cellular system  Call arrivals Poisson  Call holding and cell dwell times Scenario 1: exponentially distributed as in performance analysis Scenario 2: iid with Gamma distribution  Cell and handoff area dwell times with  = 1.5  Call holding time with  = 2 Same mean value  Cell shape: hexagonal  Each neighbor has equal probability to receive handoff

37 Simulation Results: Comparison of QoS Parameters  B OR, B ON : blocking probability of real-time & non-real-time service  P hf : Forced termination probability of real-time service calls  T N : Transmission delay of non- real-time service calls  Scen#1 and analytical analysis results are consistent < 4% difference in B OR, B ON, and P hf Accuracy of analysis is substantiated  Scen#1 and Scen#2 results are comparable P hf : Scen#2 is 20 less B OR, B ON : Scen#2 is 6% and 2% larger, respectively T N : Scen#2 is 28% less  Reasonable: Gamma has smaller standard deviation  Parallel trend: Analytical formula with tolerable error margins

38 Simulation Results: Performance Comparison of real-time calls  F hr & P hr : Priority and preemptive have 14.7% and 30.9% improvements over guard channel, respectively  B OR : almost the same  Priority especially with preemptive procedure is effective in decreasing forced terminations  Schemes: Standard guard channel (base) Priority reservation Preemptive priority handoff  Higher QoS parameters when higher arrival rates (lower service quality)

39 Simulation Results: Performance comparison of non-real-time calls  T N increases with higher traffic  Guard channel performs better Channels available for non- real-time decreases due to lower priority Largest T N is 3.91sec.; 6.5% of whole service time 31% decrease in forced termination probability is more important 7% increase in blocking probability of originating non-real-time calls Forced termination probability of non-real-time is negligibly small  Proposed scheme is better in terms of the performance

40 Conclusions  A handoff scheme is proposed Priority reservation Preemptive priority policy  Analytical model for performance analysis has been proposed  Simulation results match the analytical model  Several QoS parameters have been evaluated  Forced termination probability of handoff requests of real-time calls can be decreased  Non-real-time service handoff requests do not fail A reasonable 6.5% transmission delay increase

41 References  [1] Priority handoff analysis, Vehicular Technology Conference, 1993 IEEE 43 rd, Xie, H.; Kuek, S., Page(s): , Digital Object Identifier /VETEC  [2] CS5214 Course notes, Ing-Ray Chen, 2004.