Transcript He-JPDC09-slide - People at VT Computer Science

A proxy-based integrated cache consistency and
mobility management scheme for client-server
applications in Mobile IP systems
- Weiping He, Ing-Ray Chen
Group Members
Suresh Giridharpuram
Lakshman Krishnamurthy
Manoj Maskara
CS 5214 Spring 2010
Topics to be covered
•
•
•
•
•
•
•
•
•
•
Motivation
Suresh
Role of Proxy
Contribution of the paper
System Infrastructure
Service Handoff process
Manoj
SPN Model
Parameterization & Transition Rates
Performance Metric and Cost Functions
Performance Evaluation & Comparisons
Summary
Suresh
Lakshman
2
Motivation
• Mobile host may disconnect and reconnect
• Mobile host query dynamic data such as
stock prices, weather report, etc.
• Sending query to server and receiving
reply is expensive
• MH can cache data objects to improve
response time
• MH must ensure cache data are up-todate
3
Motivation …cont.
• Solution is Integrated Cache and Mobility
Management using proxy
– MH uses invalidation reports to determine the
validity of the cache
– If the cache is invalid query request is sent
uplink to the server
– Per user proxy to buffer invalidation
messages if MH is disconnected
– Proxy handles both service and mobility
management
4
Role of Proxy
• Serves as MH’s gateway foreign agent (GFA)
• Proxy migrates with MH when MH crosses
regional area
• Allocating extended cache space to store
service context information including cache
validation report for each MH
Goal:
Identify optimal regional service area size
to minimize network traffic
5
Contributions of the paper
(1) Integrated mobility and cache management to minimize
the overall network traffic cost for supporting mobile
client-server query applications in future MIPv6 systems
(2) Identifying the optimal proxy setting including the region
area size that will minimize the overall network traffic
generated due to mobility and cache consistency
management
(3) Benefit of integrated mobility and cache consistency
management in MIPv6 compared with basic MIPv6, noproxy and/or no-caching schemes, as well as a
decoupled scheme under which mobility management
and cache management are separate but optimally run
6
System Infrastructure
7
Service handoff
8
Algorithm
9
Algorithm…..cont.
10
SPN Model
ωw
ωs
Rate depends on the cost
of informing HA & CN and
the cost of transferring
service context of
Proxycache to new proxy
location
σ
Objective is to
find optimal K
Rate depends on the #
of subnets separating
MH & the proxy
Rate depends on the cost
of informing HA & CN and
the cost of transferring
Proxycache to new proxy
location
11
SPN Model ….cont.
12
Parameterization
13
Transition Rates
# of hops between
current subnet and
proxy
MH2Proxy Transition Rate:
# hops between 2
subnets times # packets
of service context info
MovingProxy Transition Rate:
Cost to
communicate HA
and N CNs
InquiringProxy Transition Rate:
Delay for contacting
the proxy
Moving proxy to
current AR
Inform HA and CNs
of proxy’s CoA14
change
Performance Metric and cost
function…1
Pwake =ωs / (ωs + ωw)
Ctotal = λe x Cquery + σe x Cmobility + μe x Cinvalidation
λe = λq x Pwake
σe = σ x Pwake μe = μ1+ μ2 + μ3 +…….+ μNdata
15
Performance Metric and cost
function…2
16
Performance Evaluation…1
17
Performance Evaluation …2
•
This scheme was compared with three baseline
schemes:
1) A no-proxy no-caching (NPNC) scheme,
2) A proxy no-caching (PNC) scheme, and
3) A no-proxy caching (NPC) management scheme
•
We also compare our scheme with a decoupled
scheme that optimally but separately manages
mobility and service activities.
18
Performance Evaluation …3
• The NPNC scheme essentially is the basic MIPv6
scheme without using a proxy for either mobility or cache
management.
• The PNC scheme is the proxy-based regional
registration scheme using a proxy for mobility
management.
– In these two schemes, the MH does not cache data objects.
• The NPC scheme uses basic MIPv6 for mobility
management and cached data objects maintained by the
MH for cache management, but there is no proxy used
19
Performance Evaluation …4
Optimal K to minimize
network traffic
Kopt decreases slowly as λq,i increases. When λq,i
increases, the query cost increases and the MH
prefers a small service area size to reduce the
query cost.
20
Performance Evaluation …5
Kopt increases as σ increases. when the mobility rate
is high, the proxy likes to stay at a large service area
to reduce the location handoff cost
Kopt decreases as μi increases. As μi increases, the
data in the local cache are more likely to be out-ofdate. MH sends more queries and more invalidation
reports will be sent from the CN to the MH through
the proxy. The MH will stay close to the proxy to
reduce the triangular CN-proxy-MH communication
cost.
21
Performance Evaluation …6
Kopt decreases as the MH sleeps
longer. The data in MH's local cache
are more likely to be out-of-date
when the MH sleeps longer. MH will
stay close to the proxy to reduce the
triangular CN-proxy-MH
communication cost
Kopt decreases as the number of cached data
objects increases. As the number of cached data
objects increases, more invalidation reports will be
sent from the CN to the MH. To reduce the
triangular CN-proxy-MH cost the MH tends to stay
closer to the proxy.
22
Performance Comparison …1
Caching-based schemes (NPC and
PICMM) achieve much better
performance vs. non-caching-based
schemes (NPNC and PNC), especially
when λq,i is large.
Between the two caching-based
schemes, PICMM scheme performs
consistently better than NPC due to the
use of a proxy for integrated cache and
mobility management for extra cost
saving.
23
Performance Comparison …2
PICMM scheme outperforms all other
schemes. PICMM uses a proxy to
serve as a GFA to reduce the
cost of location handoffs and the
benefit is especially pronounced when
the mobility rate of the MH is high.
PICMM
achieves much better performance
.
especially when Ndata is large because
caching saves much of the uplink cost for
query processing.
24
Performance Comparison …3
PICMM consistently performs better than NPC. However,
when the data update rate is very high, most cached data
objects are invalid, so queries will need to be routed to the
CN. In this case, there exists a cross-over point in the update
rate beyond which PICMM would perform worse than a noncaching scheme because of the CN-proxy-MH triangular cost.
PICMM incurs a higher query cost than
non-caching schemes. There is a crossover point in the sleep ratio beyond
which PICMM would perform worse than
a non-caching scheme.
25
Proxy maintenance cost
We see that the amortized cost for maintaining the per-client proxy is at its minimum when
the MH's SMR is high under which the MH moves frequently in between queries, which
increases the chance of proxy maintenance.
26
Performance Comparison with Decoupled scheme
27
Summary
• Client-side proxy with duties for both cache and mobility
management allows intelligent MHs to determine the
best service areas for service handoffs to minimize the
network traffic cost
• Computational procedure to compute the optimal service
area size to minimize overall network traffic cost
• Compared with several no-proxy and/or no-caching
schemes and a decoupled scheme and concluded that
this scheme outperforms others in terms of the network
traffic cost
• Computational procedure at static time to determine the
optimal Kopt.
• MH can perform a simple look-up operation to determine
Kopt based on data collected at runtime
28
Questions???
29