connection oriented mobility using edge point interactivity
DESCRIPTION
Connection Oriented Mobility Using Edge Point Interactivity. Sandeep Davu Networking and Media Communications Research Laboratories Computer Science Department Kent State University. AGENDA. Problem Statement Related Work Proposed Scheme -- IPMN Implementation 2-Layer IPMN - PowerPoint PPT PresentationTRANSCRIPT
Connection Oriented Mobility Using Edge Point
Interactivity
Sandeep DavuNetworking and Media Communications Research
LaboratoriesComputer Science Department
Kent State University
AGENDA
• Problem Statement
• Related Work
• Proposed Scheme -- IPMN
• Implementation 2-Layer IPMN
• Performance Analysis
• Modeling of 3-Layer IPMN
• Conclusion
AGENDA
• Problem Statement– Mobility Management– Short Comings of current network– More higher layer specific approach needed
• Related Work
• Proposed Scheme
• Performance Analysis
• Conclusion
Mobility Management • A node is said to be mobile if its point of attachment is flexible within
and across networks.• Handoff is the process of changing the point of attachment without
losing the connection states.• Handoff classifies into two categories.
– L2 handoff –only the point of attachment in LL is required (change of Access Point).
– L3 handoff—happens along with L2 handoff where Access Point is in a different subnet. This requires attaining new IP address.
Router Router
Backbone Network
Corresponding Node
Mobile Node
AP3
IP-Subnet 1 IP-Subnet 2
AP4 AP1 AP2
Mobile Node Mobile Node
L2 HandoffL3 Handoff
IP1 IP1 IP2
Short comings of network
• OS and network architecture did not envision wireless and mobility.• Error Prone nature of the wireless links – not as rigid as Ethernet.• State full nature of connection oriented networks – robust
retransmissions.• A transport layer connection is identified by a four tuple <source
address, source port, destination address, destination port> maintained at either end points.
• A L3 handoff requires a network address change for the Mobile Node rendering the current TCP connection useless.
• Unexpected disconnections and moving end points.• Rapidly Converging 3G and 4G networks and strong need to have
seamless mobility• Is there an effective way to handle this???
Protocol Re-organization
• Principle – Most network complexities are better handled at higher layers and end-systems.
• Protocol extension needed for mobility and justifies the design principle employed.
• Mobility in wireless networks (802.11 MAC) involves more than one network layer to perform a handoff.
• cross-layer interaction between networking layers to achieve high performance loss-free handoffs.
• Event based approach with no timer latencies to kick off actions.
Are current approaches not enough???
• There are several approaches proposed which address one particular area of mobility
• Solutions proposed in one layer (either IP or TCP).• Interoperability between layers was hindered.• Indirect performance implications even if the problem was
addressed directly.
AGENDA
• Problem Statement
• Related Work– Network Layer Solutions– Higher Layer Solutions
• Mobility Management• Performance Tuning
• Proposed Scheme
• Performance Analysis
• Conclusion
Network Layer Solutions
• Mobile IP provided the first effective crack in handling mobility.
• Route Optimization provides a solution for triangulation problem.
• Hint based handoffs used triggers from lower layers as hints to perform fast handoff.
• The RAT (Reverse Address Translation) architecture, based on the network address translation (NAT) protocol, uses packet re-direction service between CH and MN to support IP mobility.
Mobile IP• Added an indirection to the routing mechanism in the
form of Home Agents and Foreign Agents.• Mobile IP Handoff is a two step process
– Movement Detection—Detects which Agent will service the MN.– Registration—Registers the Foreign Agent with Home Agent.– Tunneling—HA and FA creates a mutual tunnel to route packets
to MN.
Mobile Node
Home Agent
Foreign Agent
Correspondent Node
LL
Probe SNR
LL
Assoc
MIP
Movement
MIP
TCP
AA AALT timerCOA Tunneling
LL
MIP
Tunneling IP-COA
TCP
IP
LL
Mobile NodeHome Agent
Old Foreign agent
Correspondent Host
Beacon timerBeacon Assoc
t
t
Mobile IP Movement Detection
LL
Probe Auth Assoc
LL
Auth Assoc
MIP
Registration
MIP
TCP
AA MovementCOA Tunneling
LL
MIP
Tunneling IP-COA
TCP
IP
LL
RT timer Congestion
Mobile NodeHome Agent
New Foreign agent
Corresponding Host
RT timer Congestion
Beacon
tt
Mobile IP Registration and Tunneling
Higher Level Solutions
• Split Connection Approach– I-TCP was a split approach to the end-to-end
connection where the connection was split at the Base Station or Access Point.
– MSOCKS[8] achieves connection redirection using split connection proxy.
• End-to-End Approach– TCP-R is based on an idea- same as ours, renewing
the connection to handle the new IP address.
• Performance Tuning– Freeze TCP freezes the connection during the course
of a handoff by advertising a zero window at the MN.
AGENDA
• Problem Statement• Related Work• Network Layer Solutions• Proposed Scheme
– IPMN Architecture– 2 Layer implementation (experiment and
performance)
Performance Analysis
IPMN 3-Layer Modeling
• Conclusion
Our Scheme• Uses rapid cross layer interactivity to provide high
performance connection oriented mobility support.• Based on the Interactive Transparent Networking
(InTraN) paradigm – focused on ordered cross layer interactivity, developed recently in the MediaNet Lab.
• Interactive Protocol for Mobile Networks (IPMN) allows event based access to protocol states even by network layer processes or even by L7 processes.
• This mobility solution does not require any functional change in the classical TCP/IP network, can avoid FA or HA (thus the need of an infrastructure!), can avoid triangulation, is loss-free, and above all offer much faster handoff.
Interactive Transparent Networking
6a
3b
3a
user space
1
7
TCP kernel
2
4a
Event Information
ConnectionState
Application
Probing APISubscription API
T-ware (2)
TCP Connection
system
Kernel
5
Signal Handler
4b 6b
T-ware (1)
T-ware (n)
Event Monitor
Socket API
Interactive protocol for Mobile networks
• Event based approach trapping handoff related events—mostly at L2 like probing, authentication, association.
• Probes the link layer and intelligently performs a handoff based on information from L2.
• Handoff procedure depends on the cell boundary conditions of the Access Points.– Overlapping – When a MN is being serviced by more than one
AP at any given point of time – Non-overlapping – When MN is experiencing temporary periods
of disconnections when switching between AP’s.
• Does not require an infrastructure – easy to deploy, backward compatible to legacy networks.
LL
Probe Auth Assoc
LL
Auth Assoc
IP
TCP
TCP
IP
LL
PR timer
Mobile Node
Future Access Point
Correspondent Host
0 win
Beacon
t
Application
Freeze Handler
Auth Handler
RealyIP Handler
Application
OPT=SWITCHIP
src_addr
OPT=SWITCHIP
Application
SwitchIP Handler
dst_addr
ACK ‘OPT’
Unfreeze Handler
Non 0 win
1
3
4
2
Trns timer
Handoff in an overlapping cell boundary
LL
SNR
Application
Old Access Point
t
Interactive Protocol for Mobile Networks(2- layer Implementation)
Node
Event
Layer Event tracked Action taken by event handler
Mo
bile
No
de
1 LLL2 handoff has been initiated.
Advertises a zero window to the FH. The freeze mechanism of TCP will force the FH to stop transmission.
2 IPA new IP has been assigned to the MN from the future BS.
Call the switch_ip() system call. This will replace the source IP filed in the IP header of the MN with the new IP and will send a segment to the FH with TCP option = SWITCH_IP to replace the destination IP field on the FH.
3TCP
The ‘SWITCH_IP’ segment has been ACKed.
Advertises a non-zero window to the FH. This will unfreeze the connection and enable the FH to resume transmission.
Fixe
d H
ost
4TCP
A special TCP segment received with TCP option=SWITCH_IP.
Strip the new IP number from the options part of the segment, then call the switch_IP() system call which stores the new IP in the destination IP field of the IP header overwriting the old IP number.
LL
Probe Auth Assoc
LL
Auth Assoc
IP
TCP
TCP
IP
LL
PR timer
Mobile Node
Future Access Point
Correspondent Host
0 win
Beacon
Application
Freeze Handler
Auth Handler
RealyIP Handler
Application
OPT=SWITCHIP
src_addr
OPT=SWITCHIP
Application
SwitchIP Handler
dst_addr
ACK ‘OPT’
WakeUp Handler
Non 0 win
1
3
4
5
Trns timer
Handoff in an non-overlapping cell boundary
Freeze
Pro
be B
ou
nd
ary con
ditio
n
Get N
ewIP
Sw
itch S
ou
rceIP
WakeU
p
Sw
itch
Des
tIP
New IP
Interactive Protocol for Mobile Networks(2- layer Implementation)
• Implementation of L4-L7 cross layer interactivity.• Experimented with the manipulation of IP addresses to
reflect the address change.• Implemented changes to the kernel on a FreeBSD4.5
OS running on 700MHz Intel Pentium 4 processor.• API calls for subscribing to events and t-ware
modifications to the kernel.• Experiments were carried out between different sites
(varying geographic distances)
Interactive Protocol for Mobile Networks(2- layer Implementation)
Table-2 API Extesnion
AGENDA
• Problem Statement• Related Work• Network Layer Solutions• Proposed Scheme• Performance Analysis
– Experiment Setup– Performance results
• Modeling 3-Layer Handoff.• Conclusion
Experiment Setup
BS1
MobileNode
CorrespondentNode Gateway
Switch
BS2 BS3
InternetLab setup consisted of a Mobile Node a switch and three Base station machines running FreeBSD4.5 OS and a gateway to connect to the outside world.
Handoff was simulated using the switch unplugging already plugged in BS and plugging in the new BS. MN is always connected to the switch.
Three scenarios where we performed the experiments– varying the position of the Correspondent Node each time– locally in the lab, in Texas and in Virginia.
The CN generated voice traffic based on the NetSpec Source Models . We also let the MN move along the cyclic path handoff occuring every 2 minutes BS1→BS2→BS3→BS2→BS1
Node name
Location IP numberAverage
RTT
Hops from MN
LocalKent, Ohio
131.123.36.11 1 ms 3
VirginiaChantilly
, VA66.94.95.236 90 ms 19
TexasHuston, Texas
70.241.64.99 183 ms 26
Voice
0
500
1000
1500
2000
2500
3000
1 11 21 31 41 51 61 71
Call Arrival Distribution
Call number
Interarrival time (ms)
0
5
10
15
20
25
30
1 11 21 31 41 51 61
L1 = 0.004168
L2 = 0.003334
L3 = 0.002778
Call Duration Distribution
Call number
Duration (min)
• voice has been characterized by a constant bit rate (CBR) source. Sampling rate is 8 kHz and each sample is 8 bits. This gives the standard bit rate of 64 Kb/sec for acceptable voice quality.
• Inter-arrival time between two calls is exponentially distributed.
• To generate a 64 Kb/sec voice stream, talk bursts were generated by a 144-byte blocks separated by 18 ms silence periods.
Performance Results• After running the experiment several times on the three nodes we
have observed a big difference –up to two orders of magnitude—in handoff latency between IPMN and classic MIP.
• IPMN managed to perform handoff in 110 to 200 milliseconds on average while MIP needed between 14 to 44 seconds.
• substantial reduction in handoff latency highlights the advantage of event-based protocols like IPMN over timer-based protocols like MIP.
• Demonstrates the property by comparing the handoff latencies of the first 5 handoffs at the application level and at the MIP level.
TABLE-4. HANDOFF LATENCIES (IN MS) OF THE FIRST FIVE HANDOFFS
Local Virginia Texas Handoff
IPMN MIP IPMN MIP IPMN MIP 1 106 12654 114 58669 202 51359 2 107 7124 106 24975 193 33187 3 111 1524 106 22672 195 29099 4 115 48945 111 77414 195 63523 5 109 1008 121 30772 200 41676
Average 110 14251 112 42900 197 43769
0
20
40
60
80
1 2 3 4 5
0
20
40
60
80
100
1 2 3 4 5
0
20
40
60
80
100
1 2 3 4 5
MIP level latency
Application level latency
(a) Local node
Handoff
Latency (seconds)
(b) Virginia node
Handoff
(c) Texas node
Handoff
Performance Results
• Comparing the overhead of MIP and Application.• Application cannot immediately recover as soon as handoff is
completed.• Strong Reason to have application aware network solutions for
smoother transitions.
Performance Results
0
50
100
150
200
250
300
1 5001 10001 15001 20001
IPMN MIP
(b) Virginia Node
Arrival time (seconds)
Block number
0
200
400
600
800
1 5001 10001 15001 20001
IPMN MIP
Arrival time (seconds)
Block number
(a) Texas Node
10
30
50
70
90
110
130
150
1 3001 6001 9001 12001 15001
10
30
50
70
90
110
130
150
1 3001 6001 9001 12001 15001
(a) IPMN Jitter
Block number
Interarrival time (ms)
(b) MIP Jitter
Block number
Interarrival time (ms)
Performance Results
1000
10000
100000
1000000
10000000
1 21 41 61 81 101
Alpha1=0.4Alpha2=0.55
Document Size Distribution
Document number
Document Size (bytes)
0
1000
2000
3000
4000
5000
1 21 41 61 81 101
Lam1=0.000001Lam2=0.000005
Document size distribution of the first 100 documents. Interarrival times of the first 100 documents.
xX exf )(
mean/1,
x
kxFX 1)(
21
63.04.0
k
1)( xkxf X
Performance Results
Figure-7. Traffic arrival at the MN at the two nodes for two values of λ. Handoff occurrences at both MIP and IPMN are marked on the fragment number of each plot.
Traffic arrival at MN with handoff markedLocal node - Lambda=0.00005
0200000
400000600000
800000
1 5001 10001 15001 20001 25001 30001
Message number
Arr
ival
Tim
e (m
s)
IPMNIPMN HandoffMIPMIP Handoff
Traffic Arrival at MN with handoff Local Node with λ = 0.000005
Traffic Arrival At MN with HandoffLocal node - Lambda=0.00001
0200000
400000600000
800000
1 5001 10001 15001 20001 25001 30001
Message number
Arr
ival
Tim
e (m
s)
IPMNIPMN HandoffMIPMIP Handoff
Traffic Arrival at MN with handoff Local Node with λ = 0.000001
Traffic arrival at MN with handoff markedAl-Quds node - Lambda=0.00001
0500000
10000001500000
2000000
1 2001 4001 6001 8001 10001
Message number
Arr
ival
Tim
e (m
s)
IPMNIPMN HandoffMIPMIP Handoff
Traffic Arrival at MN with handoff Al-Quds Node with λ = 0.000001
Traffic arrival at MN with handoff markedAl-Quds node - Lambda=0.00005
0500000
10000001500000
2000000
1 2001 4001 6001 8001 10001Message number
Arr
ival
tim
e (m
s)
IPMNIPMN HandoffMIPMIP Handoff
Traffic Arrival at MN with handoff Al-Quds Node with λ = 0.000005
AGENDA
• Problem Statement
• Related Work
• Network Layer Solutions
• Proposed Scheme
• Performance Analysis
• Modeling 3-Layer Handoff
• Performance Analysis
• Conclusion
IPMN 3-Layer Handoff• IPMN has L2 and L3 handoffs.• We considered 802.11 as the MAC and modeled the
handoff based on 802.11.• TCP based implementation earlier has given enough
insight to model the L3 handoff.• Closely observing we found it controlled properly there
are some phases in L2 and L3 Handoffs that could be done in parallel.
• Events from L2 triggers actions in L3.• Allowing direct access between layer would be a chaos.• Allowing application intervention is the cleanest way for
decision making and controlled cross-layer interaction.
IPMN 3-Layer Handoff• IPMN has L2 and L3 handoffs.• We considered 802.11 as the MAC and modeled the
handoff based on 802.11.• TCP based implementation earlier has given enough
insight to model the L3 handoff.• Closely observing we found it controlled properly there
are some phases in L2 and L3 Handoffs that could be done in parallel.
• Events from L2 triggers actions in L3.• Allowing direct access between layer would be a chaos.• Allowing application intervention is the cleanest way for
decision making and controlled cross-layer interaction.
Modeling 3-layer IPMN
• Divided into two layers– L2 802.11 Handoff– L3 Handoff.
• L2 handoff– Probing – Time to Scan the channels and identify a channel that
can be used.– Authentication – Once the channel is established the useability
of the channel is verified.– Association – The MN will be associated with the Access Point
when the state information is transferred from old to new access point.
• L3 handoff– Based on the previous experiments ….event notification and t-
ware overheads...
L2 handoff model for 802.11
• Link Layer Latency– Probing
– Authentication
– Association/Re-Association
eud TeTuS **
TimeMinChannelTT
TimeMaxChannelTT
de
du
*2
*2
))(*2(*))(*2(*00
TimeMinChanneldttfeTimeMaxChanneldttfuS ttd
n
iTatd unWherePdttfA
1 0
1)(
Trtd PdttfRA
0
)(*2
L2 Handoff model for 802.11
• Link Layer Latency– Probing (Scan delay)
u + e = number of channels – 802.11 has 11-16 channels
Tu : Time Spent in used channel + probe RTT
Te : Time Spent in empty channel + probe RTT
Td : Time For probe transmission
ft (t) : characteristic of the channel as a factor of time
BER … Transmission rate …
eud TeTuS **
TimeMinChannelTT
TimeMaxChannelTT
de
du
*2
*2
0
)( dttfT td
L2 Handoff model for 802.11
• Probing Delay– Total Probing delay
– Most of the Time spent in a L2 Handoff is in probing.
– There are many techniques in L2 Handoff itself to speed up Probing for overall better Handoff.
))(*2(*))(*2(*00
TimeMinChanneldttfeTimeMaxChanneldttfuS ttd
L2 Handoff model for 802.11
• Authentication– Each successfully scanned channel is tried for authentication until a
channel is authenticated.
Td : Time For probe transmission
PTa: Authentication Time
n : Total Number of tries –one per channel until authentication is succeeded.
i : Channel Number that is being authenticated
n
iTadd unWherePTA
1
1*2
L2 Handoff model for 802.11
• Association/Reassociation– After successful authentication the MN’s state information is transferred
from the old AP to new Access Point.
– Re-association is helpful in knowing the current attachment point of the MN as it is moving from old Access Point to new Access Point.
Td : Time For transmission
PTr: Time for state information exchange between Access Points after successful Authentication of that channel
Trdd PTRA *2
Modeling IPMN
• Higher Layer Latency– Signaling Overhead – Time Taken for event notification – Event Handler Overhead – Time Taken by t-ware modules to access
event information and/or any internal stored information
– Attaining IP address – Application level overhead to get the IP address
Events Tracked
Table 8: Events for IPMN
Ack for TCP_OPTION=SWITCHIP
Recv TCP_OPTION=SWITCHIPTCP Layer
IP address changeIP Layer
Successful Authentication
Successful Probing Link Layer
EventsLayer
Table 8: Events for IPMN
Ack for TCP_OPTION=SWITCHIP
Recv TCP_OPTION=SWITCHIPTCP Layer
IP address changeIP Layer
Successful Authentication
Successful Probing Link Layer
EventsLayer
Modeling IPMN
• Higher Layer Latencysignaling overhead
Event Handler overhead
Attaining IP Address
PRd: proactive registration delay for getting an IP address.
10010 XwheresXSycd
205 YwheresYTg d
dd TPR 1*2
Modeling 3-layer IPMN
• Modeled the total IPMN handoff scheme using statistical modeling.• Incorporated the L2 delay as a statistical model and all the signaling
delays and the system call delays as constants.• Also modeled Mobile IP’s handoff to have better performance
analysis.
Modeling 3-layer IPMN
MN
AP TCP LL
CH
AP TCP LL
new AP
AP LL
Pr req 1
Pr req n
Pr res 1
Pr res nE1
Freeze
Auth req
E2 Re-Asso req
Auth res
Re-Asso res
NewIP req
NewIP res
Relay IP
ACK Relay IP
Wake Up
(a) Overlapping Cell Boundary
E3
Modeling 3-layer IPMN MN
AP TCP LL
CH
AP TCP LL
new AP
AP LL
Pr req 1
Pr req n
Pr res 1
Pr res nE1
Freeze
Auth req
Re-Asso req
Auth res
Re-Asso res
Relay IP
ACK Relay IP
Wake Up
(b) Non-Overlapping Cell Boundary
E3
E5
Modeling IPMN
• Total IPMN Handoff Delay for overlapping boundary conditions
• Total IPMN Handoff Delay for non-overlapping boundary conditions
dddddddddd SycEpPRSycRATgSycATgST *3,max*2,max0
ddddddddddn SycEpPRSycRATgSycATgST *4,*2,max0
Modeling MIP
• MIP Handoff consists– Movement Detection
– Registration
– Tunneling
• MIP Handoff consists
ddd NxTxM 21 *2
2*2 dd TR
dd ITTu *2
ddddd ITRMLH
Extensions-performance analysis
• Both MIP and IPMN handoffs are simulated using the developed models.
• Performed 100 handoffs and averaged them for both overlapping and non-overlapping scenarios in either case.
• IPMN had an average handoff delay of only 70ms while MIP had an average delay of 1.37s in overlapping and 1.6s and 2.14s in non-
overlapping scenario. Handoff Latencies
IPMN
IPMN
MIP1
MIP1
MIP2
MIP2
01
23
1 2cell boundaries
tim
e (s
ecs)
IPMN
MIP1
MIP2
MIP1 and MIP2 are versions of MIP with an AA lifetime of 100ms and 1s respectively.
MIP2 is the proposed practice.
MIP1 though seems to have lower handoff delay it imposes a lot of communication overhead monopolizing the bandwidth
Extensions-performance analysis
• IPMN always stays closer to No handoff case which has delay only due to BER and congestion--Normal TCP flow if the MN were not shifting cells.
• MIP lags behind by approximately 10s delivering voice traffic and 15s in delivering WWW traffic.
• Minimal handoff transition delay for IPMN provides seamless connection.
AGENDA
• Problem Statement
• Related Work
• Network Layer Solutions
• Proposed Scheme
• Performance Analysis
• Modeling 3 layer IPMN
• Conclusion
Conclusion
• We have presented high performance mobility protocol which uses rapid cross layer interactivity.
• Eliminates routing indirection (triangulation) by explicitly specifying the application about the underlying network change.
• Flexibly manipulating the underlying network states transparently from L7 to reflect network address changes—thus eliminating the L3 movement detection.
• MIP’s timer based rediscovery of already existing state information in lower layers makes it sluggish.
• Our scheme uses this and other information from lower layers to intelligently perform handoff– proactive or reactive.
Conclusion
• Most interesting claim- solution is based on L7 disposable transientware processes.
• Demonstrated in this case one possible intelligent schema for high performance TCP/IP mobility handling.
• Further improvements and replacements by more powerful schemes are easy to incorporate– flexibility of L7 transientware.
• Demonstrated mobility solution did not require any functional change in classical TCP/IP layers.
• Performance Results indicate the effectiveness and efficiency of our even-based scheme.