Transcript LIN03-FT-mobileIP-slide

CS 6204 Paper Presentation
An Efficient Fault-Tolerant
Approach for Mobile IP in
Wireless Systems
Jenn-Wei Lin and Joseph Arul
Paper Presented by: Vidhya Dass
10/31/2006
1
Agenda
Introduction
The Proposed Approach
Fault tolerance of FA
Fault tolerance of HA
Evaluation
Analytical comparison & Simulation
Conclusion
2
Introduction
Mobile IP : Support wireless users with continuous
network connections while changing locations
Functionality of Mobile IP in wireless system
provided by:
Mobility agents in architecture of wireless systems(HA
and FA)
Drawbacks : No fault tolerance for MA failure
Approach : Resource sharing to redirect workloads
of faulty FA(HA) to other failure free FA(HA)
3
Basic mobile IP in wireless system
HA
CH
IP Network
Home network
FA
RAN
RAN
Wireless Data
serving area
RAN
MN
4
Related work :
1.MA statically equipped with one or more
redundant MA’s to work in standby or load sharing
mode
MA fails, one backup member selected as primary
mobility agent
ARP : Used to map IP address of faulty MA onto
network link layer address of selected backup
member
Disadvantages : Long registration delay since MN
registers with all MA
2.Checkpointing and logging technique : Store
mobility bindings in stable storage
5
GOAL : Provide Fault tolerance capability in wireless
system with mobile IP functionality
Fault tolerance in telecom system is “five
nines”(99.999) reliability requirement for network
design but Hardware failures follow bathtub curve.
Provide fault tolerance for MA failures.
6
Wireless data network model
System model
7
OA & M functions :
Configuration management: Configures equipment
with suitable resource parameters
Fault management: Detecting and reporting failures in
equipment
Performance management: Measures resource
utilization, loading status, concerned values in equipment
Security management: Monitors access rights to
equipment
Assumptions :
Failures only occur in MA: Detect failure by not receiving
agent advertisement messages within a time period
Fail - Stop approach: Faulty MA not send agent
advertisement messages
8
The Proposed Approach
Dynamically select multiple failure free MA as backup
set for faulty MA when a failure is detected
Workloads of faulty MA redirected to failure free MA in
backup set
Faulty FA : One or more failure free FA dynamically
selected (backup set), system initiated handoff issued
to virtually move all MN to service area of backup FA
(Continuous data executable property)
Faulty HA : One or more failure free HA dynamically
selected (backup), intercept packets moving toward
faulty HA and send them to corresponding MN
9
Fault tolerance of Foreign Agent
(FA_failure-affected MN’s) :MN in serving area and
arriving MN, cannot execute wireless data sessions
System initiated handoff to dynamically select multiple
failure free FA’s, which are backup set of faulty FA
FA_failure-affected MN’s virtually moved to serving areas of
failure free FA
Failure free FA’s adds visitor entries for FA_failure-affected
MN, that have moved into it
Informs MN’s corresponding HAs of new serving FAs and CoA
for mobility bindings
Workloads of faulty FA redirected to other failure free
FA
10
Achieve virtual movement of FA_failure-affected MN’s :
Modify RAN-FA interconnection network which is
determined by RAN’s internal FA-serving record
Initially:
FA-serving record of RAN : Identifier of fixed FA
FA Failure detected : FA-serving record of failureaffected RAN(initially served by faulty FA) reset with
identifiers of backup members
FA_failure-affected MNs served by backup members
but their location is same(still located in respective
radio coverage area)
11
RAN-FA Remapping for fault tolerance
12
Implementation of Foreign Agent
FA Failure detected : Failure event sent to OA&M fault
management
fault management initiates proposed fault tolerant
approach for FA
1.Interacts with performance management to acquire
loading status of failure free FAs, finds number of
FA_failure-affected MNs
2.Select multiple failure free FA as backup members of
faulty FA
3.Configuration management informed to configure
backup members of faulty FA by resetting appropriate
parameters to some equipment in core network &
13
update mobility bindings of MNs
Fault tolerance of Home Agent
HA functions : Mobility binding maintenance, packet
interception & packet tunneling
HA_failure-affected MN : MN’s managed by faulty HA
not able to receive packets from CHs
Select one or more failure free HA dynamically as
backup members
Mobility bindings of faulty HA restored by searching
all FA’s visitor lists
Distribute bindings to backup members: Up-to-date
location of all MN’s known from FA’s visitor list
entry(MN’s data link layer address, IP address and
home agent address)
14
Packet interception of faulty HA restored on backup
using tunneling. Routers collocated with HA on same
network segment, don’t forward packet to faulty HA
but tunnel packets to backup HA which again tunnels it
to located FAs(packet from CH to HA_failure-affected
MN sent by twice tunneling)
Packet tunneling function already present in failure
free HA
15
Packet route to HA_failure-affected MN
16
Implementation of Home Agent
Select multiple failure free HAs with low traffic(from
OA&M) as backup set and one among them, as HA’s
backup manager
Mobility binding restoration: mobility- reconstruction
message sent to each FA and responses divided by HA
backup manager based on MN’s IP address. Assigns
groups to HA backup members
Collocated routers remove routing entries of faulty HA
and add routing entries of backup members, with its
interface set to virtual interface pointing to software
program to perform packet tunneling
17
Changing the
packet interceptor
Mobility binding reconstruction
Redirecting the packet interception
18
Failure Recovery
FA recovery procedure:
Recovered FA determines failure affected RANs
(configuration management of OA&M)
Failure affected RANs reset FA-serving records to
identifier of recovered FA
Recovered FA creates visitor entries for FA-Failureaffected MNs & HAs of these MNs updated with
mobility bindings
19
HA recovery procedure:
Modify routing tables of collocated routers of
recovered HA
- packet interceptors of HA_failure-affected MN
changed
Mobility bindings of recovered HA reconstructed
- search all FAs visitor lists
20
Evaluation
Workload redirection causes
Performance degradation of failure free MA
Control message overhead
Traffic behavior of FA, HA modeled as M/G/c/c
queuing model
Assumption: Data request sent to FA & response
packet intercepted by HA follow Poisson process
Service time of data request and processing time of
packet tunneling not follow any specific distribution
21
22
Performance degradation of MA
Performance degradation of failure free FA due to
resources being contended by MN’s virtually moved
and original MN’s served
Represented as increasing blocking probability
PFA_blocking - new data request possibly blocked at
failure free FA in comparison to prefailure
Erlang’s loss formula from the M/G/c/c queuing model:
{Pre-failure}
Blocking probability
of data request to a
failure free FA
23
New blocking probability of a data request to a failure
free FA when FFA FAs fail
Blocking probability
due to original and
redirected workload
{Post-failure}
Increasing blocking probability
Post-failure
Pre-failure
24
Performance degradation of a failure free HA(HAk) Increasing blocking probability that causes an
intercepted packet to be blocked at failure free HA in
comparison with prefailure
Post-failure
Pre-failure
25
Control message overhead
Control messages issued from OA&M for assisting fault
tolerance of MA:
FA_Loading
RAN_Mapping
Binding_Update
HA_Loading
Interceptor_Change
Binding_Restoration
26
Cost of FA_Loading = TFA_Loading + TFA_Response
Transmission time from fault
management to performance
management
Transmission time from
performance management to
fault management
Cost of RAN_Mapping = TRAN_Mapping
Transmission time from configuration management to RAN(single memory
access)
27
Cost of Binding_Update =
Simultaneous transmission
of mobility binding update
command from fault
management
Total number of
FA_failure-affected MN
Total time required for failure
free FAs to send mobility
binding updates about all
FA_failure-affected MNs
Average transmission
time of registration
from FA to HA
Fraction of time
due to serial,
simultaneous
transmissions
from FAs to HAs
28
Cost of HA_Loading = THA_Loading + THA_Response
Transmission time from fault
management to performance
management
Transmission time from
performance management to
fault management
Cost of Interceptor_Change = TInterceptor_Change
Transmission time from configuration management to collocated router of
faulty HA(only memory access of routing table)
29
Cost of Binding_Restoration =
Transmission time of mobility binding
restoration to each HA from fault
management
Total number of
HA_failure-affected MN
Total time required for restoring lost
mobility binding table of faulty HA
Fraction since all FA’s
Average time of sending perform serial
simultaneous
qualified visitor entry
from FA to HA manager transmission to HA
30
manager
are negligible due to high speed physical interface of
OA&M network. Size of messages is negligible and so is cost.
where
Probability of ‘n’ in
processing data
requests/response
packets in faulty MA
Assumption : Each MN is not allowed to simultaneously issue more than one data
session . So PFA_n (P HA_n) represented as probability of n FA_failure-affected
MNs(HA_failure-affected MNs) in faulty FA(HA)
31
32
Analytical comparison & Simulation
Previous Approaches:
Primary MA has redundant MA in network segment
Co-working mode of primary and redundant MA:
Standby and Load sharing (Different performance
degradation)
Standby mode:{Pre-failure}Blocking probability of
selected redundancy 0 (No workload)
Post-failure
performance
degradation of selected
redundancy
Pre-failure
0
33
Load Sharing mode:
{Pre-failure}Load distributed among primary and all
redundancies of that primary
{Post-failure}If selected as primary then redundant MA
has to handle twice the original load
Post-failure
Pre-failure
Where (1+RAgent) is primary MA+ Redundant MA’s in network segment
Arrival rate of data to MA based on load sharing mode
34
35
Tolerate N-1 failures in N MA system
Improvement
least
Traffic intensity of MA: Expected number of arrivals per
mean service time at a MA
Performance degradation of MA : Increasing blocking
probability of failure free MA
36
Proposed Approach : Workload of faulty MA evenly
redirected to all failure free MA
Ratio of redirecting workloads to failure free MA
Previous Approach : Load sharing mode: Number of
redundancies in network segment affects increasing
blocking probability
37
38
CAgent=50
Previous Approach
One redundant MA take over one
faulty MA. Independent of Fagent
Increasing blocking probability 
as Agent/Agent 
Current Approach
Increasing blocking prob  as
FAgent 
Increasing blocking probability
not always  as
Agent/Agent 
CAgent=50
Total number of resource units in FA
and HA is 50
Average number of in processing data
requests in an FA(HA) cannot be greater
than 50.
Equal
Maximum number of NFA_MN(N HA_MN) is
50
Conclusion : Overhead of mobility
binding update is restricted by the total
resources in FA(HA)
Failure recovery overhead depends on
this graph
NFA_MN & NHA_MN under one faulty FA (HA)
39
Simulated using NS-2
Below 10%
Same workloads to an FA and HA then PFA_Blocking and
PHA_Blocking should be equal but simulation doesn’t
agree???
Stopped with 8 faulty MA’s???
Why didn’t they consider all the cases for Number of faulty MA???
Difference rate is varying randomly???
Only considering MN in data session for FA_failure-affected MN not
justified
40
Conclusion
Utilizes available resources in other failure free MA to
dynamically generate backup set for each faulty MA
Advantages :
 No Hardware support required
No failure free overhead
Distribute fault tolerant overhead to avoid significant
performance degradation on single failure free MA
Good when
is 200 and FAgent is small
41
Thank you
Questions