A Comparative Cost Analysis of Degradable Location Management Algorithms in Wireless Networks
-
Upload
ethan-goff -
Category
Documents
-
view
21 -
download
0
description
Transcript of A Comparative Cost Analysis of Degradable Location Management Algorithms in Wireless Networks
A Comparative Cost Analysis of A Comparative Cost Analysis of Degradable Location Management Degradable Location Management Algorithms in Wireless NetworksAlgorithms in Wireless Networks
Ing-Ray Chen and Baoshan Gu
Presented by: Hongqiang Yang , Jianghui Ying
Northern Virginia Center, Virginia Tech
Agenda Agenda
Architecture of Paper:– Introduction– Notion of Degradable Location Management Algorithm – Examples of Degradable Location Management
Algorithm– Two-tier Hierarchical Framework– Comparison– Application in the analysis of service handoffs– Summary
IntroductionIntroduction
ProblemA class of degradable location management
algorithms in Personal Communication Service network for tracking mobile user in two-tier HLR-VLR structure
IS-41, FRA, PLA, LAAWhich is better?
IntroductionIntroduction
Objective– Develop a uniform framework to provide a cost analysis of
location update and search operations for a class of degradable location management algorithms in personal communication service (PCS) networks
IntroductionIntroduction
Method--Two level hierarchical modeling framework High-level model Calculate the costs incurred to the PCS network:
Location-update operations Call-delivery operations
Low-level model A stochastic model to estimate the values of high-level model
parameters
IntroductionIntroduction
Location Management Scheme– Aimed to minimize the total cost Location Update
Occurs when a mobile user moves to a new location
Call delivery Occurs when there is a call for mobile user and the network must
deliver the call.
IntroductionIntroduction
Workload condition is indicated by Call-to-Mobility ratio (CMR) – CMR is high: Location Cache Scheme is
effective
– CMR is low: FRA, PLA, LAA are effective
Degradable Location ManagementDegradable Location Management
No assumption regarding the structure of the PCS network
Conceptually, HLR is in high level and VLR is in low level.
Maybe some network switches connecting the HLR to VLRs in the mobile network
Degradable Location ManagementDegradable Location Management
IS41:– Service area divided into registration areas each
corresponding to a VLR
User moves to a new registration area
Send the registration information to the new VLR
Update information
A hierarchical PCS network
HLR
PSTN
RSTP
LSTP LSTP LSTP
VLR VLR VLR VLR VLR VLR
… … …
Home Location Register
Public Switched Telephone Network
Regional Signaling Transfer Point
Local Signaling Transfer Point
Visitor Location Register
Degradable Location ManagementDegradable Location Management
The state of a location management algorithm: depends on the extent to which the location information has been degraded since the last location update operation was performed to the HLR– Strong– Weak
Degradable Location ManagementDegradable Location Management
Strong State:– means the HLR store the current information of the
mobile user in its own database– HLR can find the mobile user directly
Weak State:– means the current information of the mobile user is
not in the HLR– HLR must consult location database stored
elsewhere to find the mobile user
Degradable Location ManagementDegradable Location Management
The spectrum of degradable location algorithms– encompasses all existing location management
algorithm– differ on how fast and in what way the system’s
state degrades over time as the user moves across VLR boundaries
Degradable Location ManagementDegradable Location Management
The spectrum of degradable location algorithmIS-41
LAA(2)
FRA(2)
PLA(2)
LAA(3)
FRA(4)
PLA(3)
Paging
Degradable Location ManagementDegradable Location ManagementIS-41
User moves across VLR boundary
Send information to new VLR
Update information in HLR
Degradable Location ManagementDegradable Location ManagementFRA(K)
User moves across VLR boundary– Length of link of pointers < K:
Set up a pointer to the two involved VLRs
– Length of link of pointers = K:Update information in HLR
Degrades over time as more and more moves
Degradable Location ManagementDegradable Location Management
PLA(n)– HLR records a VLR as the agent– The agent covers all VLRs within a distance n from the
agent( local region)– No update at all with local moving– Update when across local region boundary– Paging starts from the agent
Degrades over time since the location of the mobile user becomes fuzzier to the HLR as more and more time has elapsed since the last update performed
Degradable Location ManagementDegradable Location Management
LAA(n)– HLR records a VLR as the local anchor(LA)– The LA covers all VLRs within a distance n from the
LA– Update in LA with local moving– Update in HLR when across a regional switch boundary
Three steps researching:– HLR -> LA -> VLR
Degradable Location ManagementDegradable Location Management
Performance metric:– The network cost due to location update
between two successive calls to a mobile user
Paper contribution: – Provides method to quantifying the location
update cost for a given location management algorithm with respect to the basic scheme
– Classify the algorithm
0
0.2
0.4
0.6
0.8
1
Location Update Cost:CMR=0.1
IS-41
LAA(2)
FRA(2)
PLA(2)
LAA(3)
FRA(4)
PLA(3)
Paging
The ratio of the average communication cost between two VLRs to the average communication cost between the HLR and a VLR is 0.3
Degradable Location ManagementDegradable Location Management
Classify mobile users into priority classes based on their quality of service(QoS) requirements
Separate QoS classes are being served by separate location management schemes
Simplify the per-user-based location management to the per-class-based location management
Modeling Degradable Location Modeling Degradable Location Management AlgorithmManagement Algorithm
Description of two-level modelExamples of Degradable location
management algorithm– Description of algorithm– Low-level model
Modeling Degradable Location Modeling Degradable Location Management AlgorithmManagement Algorithm
Two-level model– High-level model: cost model
Update cost Xupdate
Query cost Xsearch
Xcost = Xupdate * / + Xsearch
– Low-level model Aimed to provide the parameters for high-level model
– Low-level model Parameterize equation in high-level model Define parameters in our model
Per-mobile-user parameters
The rate at which the mobile user moves across VLR boundaries
The rate at which the mobile user is being called
CMR / , the call-to-mobility ratio of the mobile user
Mobile network parameters
T The average VLR-HLR round-trip communication cost
The average VLR-VLR round-trip communication cost
Basic HLR/VLRBasic HLR/VLR
A mobile user is permanently registered under a location register(HLR)
Mobile user enters a new VLR areaReports to VLRVLR inform HLRHLR update location information
Location Update Cost = 1
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
Agent: the VLR performs the last update operation to the HLR
HLR update location information only when the distance between the agent and the current VLR is great than or equal to a predefined distance value n
R0R1
R1R1
R1
R1R2
R2R2
R2R2
R1
R2
R2
R2R2
R2R2
R2
SDF partitioning under the hexagonal model
PSTN
HLR
F
A
Local Movement: within the 2-Distance region
Update Action: None
Regional Movement: Outside of 2-Distance Region
Action: Update the HLR
PLA Algorithm
ED
CB
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA) Modeling parameters
– n: specify the (n-1)-distance region within a user causes no update cost
i : the mobility rate of the mobile user moving from ring i to ring i + 1
i : the mobility rate of the mobile user moving from ring i + 1 to ring i
r : the execution rate to perform a location update to the HLR
I : the execution rate to locate the mobile user currently located in ring I
– P(i,j):The probability that the system is in a particular state in equilibrium
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
Assuming random move The probability of the mobile user moving from
ring i to i + 1, i >= 0(2i + 1) / 6i , if i >= 1,
1, if i = 0. The probability of the mobile user moving from
ring i + 1 to i, i >= 0(2 (i + 1) - 1 )/ (6(i +1)), if i >= 0,
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
The mobility rate of the mobile user moving from ring i to ring i + 1
(2i + 1) / 6i , if i >= 1,
i = , if i = 0.
The mobility rate of the mobile user moving from ring i + 1 to ring i i = (2i +1) /( 6* (i + 1)) , i >= 0
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
The execution rate to perform a location update from a new VLR agent to the HLRr = 1 / T
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
The time to locate the mobile user in ring i– From HLR to agent– From the agent to the current VLR(in ring i)– From the current VLR back to the HLR
The execution rate to locate the mobile user in ring ii = 1/(T + 0.5*(3i2 + 3i) * )
Query one-half of the VLRs in the i-distance region
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
Low-level Makov model – State represented by (a, b)
a: 0, IDLE 1, CALLED
b: the current distance between the mobile user and the local agent
Markov Model for PLAMarkov Model for PLA
0,0 0,1 0,2 0,n-1 0,n-1*
1,0 1,1 1,2 1,n-1 1,n-1*
r
0 1
n-1
0 1 2 0
2
01 2 n-1
0 1 21 2 n-1 r
Average cost for location update operation
Average cost for location search operation
Average total cost:
Paging and Location Algorithm(PLA)Paging and Location Algorithm(PLA)
))(1/ ) P (P( pla1
0
*1)-n(i,1)-n(i, i
update
) (1/)P(P
) ) (1/ P(pla
01)*-n(1,*1)-n(0,
j
1
0i
1-n
0j
j)(i, search
searchupdate plaplapla /cost
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
HLR only points to the VLR at the beginning of the forwarding chain
User moves across VLR boundary– Length of link of pointers < K:
Set up a pointer to the two involved VLRs
– Length of link of pointers = K:Update information in HLR
PSTN
HLR
F
A
Local Movement: length of the forwarding chain is less than 5
Update Action: update the Forwarding Pointer between Two involved VLRs
Regional Movement: Chain Length = 5
Action: Update the HLR
FRA Algorithm: K=5
ED
CB
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
Two additional behaviors in modified Markov Model under FRA– The forwarding chain will be reset after a
location query operation is performed– When the mobile user moves back to the
previously visited VLR in the chain, the length of the forwarding chain is reduced by 1 and no pointer deletion operation is required between
Obsolete pointers will be purged automatically after a period of time much greater than the average reset period
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
Looping behavior is accounted in Markov model
Assume : when the mobile user moves across a VLR, the forwarding pointer information will be updated before it crosses another VLR– Imply that the dwell time of mobile user in a
VLR is much greater than the forwarding pointer update operation time
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
Model Parameters – k : the length of the forwarding chain at which a reset
operation is performed n : the mobility rate of the mobile user moving to a new
VLR. b : the mobility rate of the mobile user moving to the
previous VLR r : the execution rate to reset a forwarding chain, i.e. to
update the HLR f : the execution rate to set-up or travel a pointer
between two VLRs i : 0<= i <=k-1, the execution rate to locate the mobile
when the length of forwarding pointer is i.
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
The mobility rate of the mobile user moving to a new VLRn= (5/6) *
The mobility rate of the mobile user moving to the previous VLRb= (1/6) *
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
The execution rate to reset a forwarding chain, i.e. to update the HLRr = 1/T
The execution rate to set-up a forwarding chain, i.e. to update the HLR
f = 1/ The execution rate to locate the mobile when the
length of forwarding pointer is i( 0<= i <= k-1)
i = 1/(T + i *)
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
Low-level Markov Model– States represented by (s1, s2)
s1= 0 , standing for IDLE
1 , standing for Called s2 indicates the current length of the forwarding chain
– States followed by symbol ‘*’ means the mobile user just enters a new VLR but the forwarding pointer operation is not yet performed
Markov model for the PCS network under FRAMarkov model for the PCS network under FRA
0,0 0,1* 0,1 0,2* 0,k-1 0,k-1*
1,0 1,1* 1,1 1,2* 1,k-1 1,k-1*
0
1 k-1
b b
b b
f n f
f n f
n
r
n
r
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
P(i , j) : represent the probability that the system is in state (i , j)
fraupdate: average cost to perform a location update operation
)/1(
))()((
)/1(
)(
1
1
)*,1(*),0(),1(
2
0
),0(
)*,1(*),0()1,1()1,0(
f
k
i
iii
k
i
i
r
kkkkupdate
PPPP
PPPPfra
Forwarding and Resetting Algorithm(FRA)Forwarding and Resetting Algorithm(FRA)
frasearch: average cost to perform a location search operation
fracost: the average total cost
))/1()((1
0
)*,1(*),0(),1(),0( i
k
i
iiiisearch PPPPfra
searchupdate frafrafra /cost
Local Anchor Algorithm(LAA)Local Anchor Algorithm(LAA)
Basic idea Location Registration operations should be as localized
as possible so as to reduce the number of registration messages to the HLR.
Local Anchor (LA)The VLR which performs the last registration operation
with the HLR is called the LA of the mobile user.
Local Anchor Algorithm(LAA)Local Anchor Algorithm(LAA)
LA’s Coverage
N M
(VLR in ring)
Coverage
1 1 1
2 6 7
3 12 19
… … …
n 6(n-1) 3n2 – 3n + 1
Local Anchor Algorithm(LAA)Local Anchor Algorithm(LAA)
AlgorithmMovement: Cross a VLR boundary but within the local
region
Operation: New VLR informs the LA without informing the HLR
Movement: regional move
Operation: New VLR informs the HLR and also becomes the new LA of the mobile
user.
PSTN
HLR
F
Local Movement: within the 3-Layer Area
Update Action: Only Update the LA(VLR A)
Regional Movement: Outside of a 3-Layer Area
Action: Update the HLR
LAA Algorithm
E
DCBA
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Parameters
n: specify the n-layer VLR region which covers 3n2 – 3n + 1 VLRs
P1: the probability of the mobile user moving under the same network switch
P1 = 6*(3n2 – 3n + 2) – (12n - 6)/6*(3n2 – 3n + 1)
= (3n2 – 5n + 2) /(3n2 – 3n + 1)
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Parameters
σ1: the mobility rate of the mobile user moving under the same network switch
i.e. σ1 = P1σ
= (3n2 – 5n + 2) /(3n2 – 3n + 1)σ
σr: the mobility rate of the mobile user crossing a network switch boundary
i.e. σr = (1- P1)σ
= (2n - 1) /(3n2 – 3n + 1)σ
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Parameters
μq: the search execution rate when the mobile user is located in the agent’s area
μq = 1/T
μa: the search execution rate when the mobile user is not located in the agent’s area
μq = 1/(T + τ)
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Parameters
δ1: the execution rate to update the agent, i.e. to set up a pointer between the new VLP and the agent
δ1 = 1/ τ
δ: the execution rate to update the HLR
δ = 1/ T
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Markov model state representation (a, b, c)
a: whether the mobile user is in the state of being called
0 ---- idle 1---- busy
b: whether the mobile user makes a move
0 ---- not move 1---- local move
2 ---- regional move
c: whether the agent currently covers the mobile user
0 ---- yes 1 ---- no
0,0,0 0,1,10,0,10,1,0
1,0,0 1,1,11,0,11,1,0
0,2,1 1,2,1
λ λ λ λμg
μa
σr
σr
σr
σr
δ δσ1
σ1
σ1σ1
δ1
δ1
δ1
δ1
λ
(0, i, j) (1, i, j) when a call comes, the transition rate is λ
(1, 0, 0) (0, 0, 0) with the transition rate of μq
(1, 0, 1) (0, 0, 0) with the transition rate of μa
Regional move(i, 0, j) (i, 2, 1) with the transition rate of σr
(i, 2, 1) (i, 0, 0) with the transition rate δ
Local move(i, 0, j) (i, 1, j) with the transition rate of σl
(i, 1, j) (i, 0, 1) with the transition rate δl
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Markov model Laaupdate
)/1()(
)/1(
)]/1()1()/1([
)1,2,()0,0,(
1
1
0
1
0
111
1
0
1
0
),1,(
),0,(
ii
i j
i j
update
PP
P
PPP
laa
ji
ji
Modeling a PCS network Modeling a PCS network operating under LAAoperating under LAA
Markov model Laasearch
Laacost
)/1()(
)/1(
1
0
)1,1,()1,0,()0,1,(
1
0
)1,2,()0,0,(
a
i
iii
g
i
ii
search
PPP
PP
laa
searchupdatet laalaalaa /cos
Analysis and ComparisonAnalysis and Comparison
Using two two-level hierarchical modeling Compare PLA, FRA, LAA with basic HLR/VLR IS-41 algorithm
VLR-to-VLR/VLR-to-HLR = 0.3
T = 1 τ = 0.3
IS-41
where IS-41update = T
IS-41search = T
searchupdatet ISISIS 41/4141cos
Comparison of PLA under different n-distance values
Performance n=3 is better than n=2 when CMR is small
Larger n means local agent can cover larger area, thus a smaller probability to cross a regional boundary. Consequently the number of update operations to HLR is reduced as n increase.
After CMR exceed a threshold, n=2 is better than n=3
As CMR increase, the larger location query cost attributed by the larger cover area starts to dominate the reduced location update cost.
searchupdatet plaplapla /cos
threshold
Comparison of FRA under different k values
Performance k=4 is better than k=2 when CMR is small
Location update cost dominates the location query cost, so a longer chain is favored since it reduces the location update cost .
After CMR exceed a threshold, k=2 is better than k=3
As CMR increase, higher cost associated with location update operations which happen frequently starts to offset the benefit of lower update operation cost.
threshold
searchupdatet frafrafra /cos
Comparison of LAA under different distance values
Performance n=3 is better than n=2 when CMR is small
Larger n means local agent can cover larger area, thus a smaller probability to cross a regional boundary. Consequently the number of update operations to HLR is reduced as n increase.
After CMR exceed a threshold, n=2 is better than n=3
As CMR increase, the larger location query cost attributed by the larger cover area starts to dominate the reduced location update cost.
threshold
searchupdatet laalaalaa /cos
Comparison of PLA, FRA and LAA head to head Comparison of location
update cost only Provide the basis for classifying
existing degradable location management algorithms based on the update cost per move relative to IS-41 for a wide range of CMR.
IS-41 keeps the system in Strong State all the time, LAA(2) is next to it in terms of maintaining the location information in a good state.
PLA(3) incurs the least amount of update overhead, since under PLA(3) the possibility of local movement is high.
PLA(3) least
LAA(2) highest
Comparison of PLA, FRA and LAA head to headComparison of the search
cost only PLA(3) incurs the most
overhead to deliver a call
Comparison of the search cost onlyComparison of location update cost only
LAA 2FRA 2PLA 2LAA 3FRA 4PLA 3
PLA 3PLA 2FRA 4LAA 3FRA 2LAA 2
Comparison of total communication cost
PLA(3) is the best when CMR is 0.1
FRA(4) is the best when CMR>0.3
PLA(2) is the worst when CMR is small
PLA(3) is the worst when CMR is large
Comparison of total communication cost for multiple users
All users are served under a single algorithm against the case when individual users are served by their respective per-user best algorithms.
As the number of users increases, the total cost difference increases because of cumulative effect.
Under single algorithm, FRA(4) is the best.
Per-user selective strategy outperform all single-algorithm cases.
Service HandoffService Handoff
What is Service HandoffThe process of transferring or migrating the service of a client from one server to another in client-server applications
Difference between location handoff and service handoff transfer of context information
Static: user profile, authentication data Dynamic: files opened, locks, stamps
Service HandoffService Handoff
Analysis of Service Handoff Using LAA scheme Assumptions
A service area corresponds to a network switch area, when a user moves across a switch boundary a service handoff will be triggered.
Context transfer communication cost Cs
Mobile user communicates with the serer by means of operations
The location and service networks are integrated
Service HandoffService Handoff
Cost factors The communication cost between the server and the LA,
the cost is τ The communication cost between the LA and the VLR in
which the mobile user currently resides if the LA is not the current VLR, which is also τ
The communication cost of migrating the service context from the old server to the new server if during the time of access, the mobile user happens to cross a service boundary, the cost of which is Cs
Service HandoffService Handoff
Recall Markov model of LAAstate representation (a, b, c)
a: whether the mobile user is in the state of being called
0 ---- idle 1---- busyb: whether the mobile user
makes a move 0 ---- not move 1---- local move 2 ---- regional movec: whether the agent
currently covers the mobile user 0 ---- yes 1 ---- no
Service HandoffService Handoff
Average cost per service operation Reward of Cs+τ to states in
which component b is 2, i.e. (0, 2, 1) and (1, 2, 1)
Reward of τ to states in which c is 0, i.e. (0,0,0) (1,0,0) (0,1,0) and (1,1,0)
Reward of 2τ to states in which c is 1 but b is not equal to 2, i.e. (0,0,1) (1,0,1) (0,1,1) and (1,1,1)
Crossover Point
Cost per service operation in LAA under different CMR and Cs values
τ =1 When Cs is relatively low, LAA(2)
is better especially at low CMR value.
Since when CMR when CMR is low and the probability of crossing switch boundary is low, so the call ratio must be low, therefore the contribution of context transfer cost is low for both schemes. There are more VLRs covered when n=3, the contribution of the 2rd cost factor is higher in LAA with n=3.
Cross point shift left as CMR increases.
As Cs increases, LAA(3) will be favored over LAA(2) for smaller probability of crossing boundary
As CMR increases, the advantage become manifest even in small Cs.
SummarySummary
Discussed the notion of degradable location management algorithm used in PCS for locating users
Classified existing location management algorithmsDeveloped methods to obtain the location update cost
for location management algorithmsTested the method by modeling PLA, FRA and LAA
and demonstrated the applicability of the modeling method
Showed how the modeling methodology can be applied to support service handoff
Conclusion and Future WorkConclusion and Future Work
Future WorkIntroducing a real-time component into the design
and deriving conditions under which user location queries can be satisfied in real-time while minimizing the location update cost
Considering users with different priority classes and discovering an optimal way to design location management algorithms so that a global design objective can be best satisfied
Investigating the applicability of the uniform framework developed in this paper to the analysis of tree-based location management algorithms.
Thank You!Thank You!