He07-slide - People

Download Report

Transcript He07-slide - People

A Proxy-Based Integrated Cache
Consistency and Mobility
Management Scheme for Mobile
IP Systems
Weiping He, Ing-Ray Chen,and Baoshan Gu
Department of Computer Science, Virginia Tech
Presented by: Jing (Julia) Xu
Apr 11, 2007
Outline







Introduction
System Infrastructure
Cache Consistency
Mobility Management
Performance Analysis: SPN
Experiment Results
Conclusion and Future Work
2
Introduction

Support mobile clientserver applications.



Query dynamic data
Reduce access cost
Improve response time
3
Introduction (Cont’d)

Challenge: Maintain service continuity with
disconnection tolerance.

Disconnection:



Voluntary: connection cost, power consumption.
Involuntary: handoff, wireless link failure.
Service continuity: be able to reuse valid previous
data when mobile host (MH) reconnects.
4
System Infrastructure
5
Cache Consistency

Overview:



MH can cache data
objects locally.
Proxy-based cache:
invalidation message.
Up-to-date data in
MHCache reduce query
traffic and cost.
ProxyCache
reconnect
disconnect
MHCache
query
data
invalidation
6
Mobility Management

To further reduce the network traffic: mobility
management by the proxy.



The proxy also serves as the Gateway Foreign Agent
(GFA) to maintain the location information of the MH.
When the MH moves across a subnet boundary within
the service (GFA) area, it obtains a new care of
address (CoA). The proxy is informed of CoA change,
but not HA or CNs.
When MH moves across a service (GFA) area, a new
GFA becomes proxy and obtains ProxyCache. HA and
CNs are informed.
7
A Trade-off

A large service area:



The proxy will not move often.
The query and cache invalidation cost is high because
of the larger distance.
A small service area:

The proxy moves often:




moving proxy cache
informing HA and CNs
The query and cache invalidation cost are lower
because of the shorter distance.
Therefore an optimal service area size exists.
8
Performance Analysis



A performance model based on Stochastic Petri
Nets.
Objective: to derive an equation to calculate the
overall network traffic cost as a function of the
number of subnets covered in a service area.
The MH can determine the dynamic optimal
service area size at runtime.
9
Performance Analysis (Cont’d)


Xs: # of subnets crossed by the MH since the MH enters a new
service area
Compute the optimal number of subnets by varying K.
10
Performance Analysis (Cont’d)

Transition rates:






Moving = 
Wake2Sleep = ww
Sleep2Wake = ws
MovingProxy =
MH2Proxy =
InquiryProxy =
K: subnet crossings
F(k): number of hops
:one-hop communication delay
per packet in the wired network
: ratio of the communication
delay in the wireless to the
wired
: average distance between
proxy and HA
: average distance between
proxy and a CN
N: # of CNs the MH engages
nCT: # of packets during a
proxy transfer
11
Performance Analysis (Cont’d)

Cost functions:

a

Ci,query =

Ci,mobility =

Ci,invalidation =
K: subnet crossings
F(k): number of hops
:one-hop communication delay
per packet in the wired network
: ratio of the communication
delay in the wireless to the
wired
: average distance between
proxy and HA
: average distance between
proxy and a CN
N: # of CNs the MH engages
nCT: # of packets during a
proxy transfer
nD: # of packets to hold a data
object
12
Experiment Results
There exists an optimal proxy service area size Kopt
13
Experiments (Cont’d)
Kopt exists for all q, i
14
Experiments (Cont’d)
Kopt increases as  increases
15
Experiments (Cont’d)
Kopt decreases as i increases
16
Experiments (Cont’d)
17
Experiments (Cont’d)
18
Experiments (Cont’d)
19
Experiments (Cont’d)
20
Experiments (Cont’d)
21
Conclusion and Future Work



This scheme allows MH to dynamically
determine the best service area size to minimize
the overall network traffic generated.
Per MH savings can have significant impacts on
all mobile users over a long time period.
Future work:


To investigate applications to which this scheme can
apply.
To investigate conditions under which it’s more
beneficial for CN to send data instead of invalidation.
22

Thank you! Questions?
23