routing over ip (trip) performance evaluation of telephony
TRANSCRIPT
University of Kansas
Performance Evaluation of Telephony Routing over IP (TRIP)
Matthew C. SchlesenerMaster’s Thesis Defense
November 25, 2002
Defense Committee:Dr. Victor Frost (Chair)Dr. Joseph EvansDr. Gary Minden
University of Kansas2
Presentation Organization� Introduction�Background�TRIP Research Issues�TRIP Model Description�TRIP Simulation Results & Conclusions� Summary of Conclusions�Next Steps�Acknowledgements�Questions
University of Kansas3
Introduction� Telephony Routing over IP (TRIP) is new signaling protocol
being developed for use in an IP network.
� The most basic function of TRIP is to locate the optimum gateway out of a Voice over IP (VoIP) network into the Public Switched Telephone Network (PSTN).
� Thesis investigation will center on the impact of varying physical characteristics of a TRIP-enabled system such as traffic load and network topologies via changes in propagation delay.
University of Kansas4
Background�Control Signaling:
� Exchange of messages related to call setup, monitoring, teardown, and network management information.
� Provides command and control infrastructure for communications networks.
� Signaling System 7� Predominant control signaling network for PSTN.� Signaling Point: use signaling to transmit and receive control
information .� Signaling Link: interconnect signaling points.� Signaling Transfer Point: transfer signaling messages from one link to
another.� Signaling Control Point: database for SS7 network.
University of Kansas5
Background� SS7 Network
� Figure from: International Engineering Consortium, http://www.iec.org/online/tutorials/ip_in/topic01.html, 2002
User A
User C
User B
University of Kansas6
Background� Signal Transport (SigTran)
� Developed to allow VoIP networks to utilize the extensive functionality and superior performance of SS7.
� Interworks VoIP network with SS7/PSTN� SS7 packets are encapsulated in IP packets by Signaling GW
and sent to Media GW Controller which makes routing decisions.
� Media stream (voice) is encapsulated in IP packets by Media GW.
University of Kansas7
Background�Resource ReSerVation Protocol (RSVP)
� Designed to provide integrated services across the Internet.� Host requests service with very specific connection parameters from
the network. � Network routers along the specified path will each be requested for
dedicated resources (e.g., bandwidth).� If all nodes along the path dedicate the resources, the reservation is
complete and the host may begin use.
RVSP Request
RVSP Request
RVSP Request
RVSP Request
University of Kansas8
Background�Voice over IP (VoIP)
� A network that transmits voice packets over IP.� Voice signal is digitized, compressed and converted to IP
packets.� Specialized signaling protocols are used to set up and tear
down calls, carry information required to locate users and negotiate capabilities.
� Examples are H.323 and SIP.
University of Kansas9
Background� Session Initiation Protocol (SIP)
� VoIP signaling protocol.� Begins, changes and terminates network sessions.� Provides advanced signaling and control to an IP network.� User Agent: end users of the SIP network that initiate requests and are
the destination of services offered across the SIP network.� Registrar: manage user agents assigned to their network domain.� Proxy Server: forward SIP requests and responses.� Redirect Server: take SIP requests and return location information of
another user agent or server.� Location Server: locates the next-hop for an incoming session request.� Also, media GW and signaling GW for interworking with PSTN.
University of Kansas10
Background� SIP Network
Proxy/Registrar/Redirect ServicesLocation Server
ENTERPRISE IPNETWORK
PRIIP IP
1 2 3
4 5 67 8 9
* 8 #
PSTN/SS7
Signaling/Media GWSignaling/Media GW
IP Network
PRI
SIP SIP
1 2 3
4 5 67 8 9
* 8 #
IP
SIP
SIP PhoneUser Agent
Soft PhoneUser Agent
SIP
SIP
SIPIP
IP
Dashed Lines areSignaling links
User A
University of Kansas11
TRIP� The most basic function of TRIP is to locate the optimum
gateway out of a Voice over IP (VoIP) network into the Public Switched Telephone Network (PSTN).
� TRIP is a voice routing protocol that is independent of the underlying VoIP signaling protocol.
� Common TRIP implementations use either SIP or H.323.� Benefits of a TRIP-enabled Network
� Reachable routes must be manually provisioned in the GW only.� Location server has knowledge of dynamic state of gateway
resources.� Dynamic routing which provides the optimum path for session
routing.
University of Kansas12
Location Server (LS)� Exchanges route table information with other location
servers.� TRIP builds a routing table for the LS it supports. � The LS will utilize that routing table to make session request
forwarding decisions.� Call requests can be rerouted between location servers based
on dynamic state of gateway resources.� A location server running TRIP is referred to as a TRIP
speaker.
University of Kansas13
TRIP-lite (TRIP-GW)� TRIP-lite advertises route
and prefix destinations to at least one location server.
� Attributes advertised include destination prefixes, capacity to each prefix destination, dynamic utilization of each trunk group.
� TRIP-lite allows the LS to have real-time knowledge of available GW resources.
P R I
P S T N /S S 7
L it e
P R I
L S
P R I
T R IP - l i teM e s s a g in g
L i t e
L i te
S IP P r o x y
G W 1
G W 2
G W 3
P r e f ix 9 1 3 & 8 1 6P r e f ix 9 1 3
University of Kansas14
I-TRIP� Interior Administrative
Domain Routing (I-TRIP)� I-TRIP exchanges GW
location and routing information insidethe locally administered domain(ITAD).
� I-TRIP database update messaging is flooded to all LS’s throughout the ITAD, based on OSPF.
Lite
LS
SIP Proxy
TRIP UpdateFlooding
ITAD
PRI
SS7/PSTN
Lite
Lite
LS
PRI
PRI
LS
Lite
PRI
TRIP-liteMessaging
TRIP-liteMessaging
SIP Proxy
SIP Proxy
University of Kansas15
E-TRIP�Exterior Administrative Domain Routing (E-TRIP)�E-TRIP exchanges telephony routing information
between administrative domains. �Based on
BGP.
LS
Lite
Lite
LS
SIP Proxy
SIP Proxy
LSLS
Lite
Lite
SIP ProxySIP Proxy
I-TRIPI-TRIP
E-TRIP
ITAD BITAD A
TRIP-litemessaging
PRI PSTNPRI
TRIP-litemessaging
University of Kansas16
TRIP Research Issues� What is the impact of LS-to-GW and LS-to-LS propagation
delay on blocking probability in a TRIP environment?� The importance of this issue to a carrier concerns deployment of
location servers to support the network. Determining propagation delay impact on call blocking will allow designers to place the fewest number of LS to support demand in a given geographic area.
� What is the impact of LS-to-GW and LS-to-LS delay on call request delivery in a TRIP environment?� The importance of this issue to a carrier concerns quality of service
provided to customers. Determining where delay will be incurred will allow designers to meet delay budget.
University of Kansas17
TRIP Research Issues� What is the impact of LS-to-GW and LS-to-LS delay on
location server call request rerouting in a TRIP environment?� The importance of this issue to a carrier concerns call setup.
Determining the impact of call reroutes on call setup delay impacts the delay budget.
� What is the impact of traffic intensity along with LS-to-GW and LS-to-LS delay variation in a TRIP environment?� The importance of this issue to a carrier is to understand how the
network will react under varying load conditions. A network designer would be able to design the network to a specific load.
University of Kansas18
TRIP Research Issues�What is the impact of trunk failure along with LS-to-
GW delay and LS-to-LS delay variation on blocking probability in a TRIP environment?� The importance to a carrier is to understand how a failure will impact
customers.
�How does the system blocking performance of a TRIP network differ from the blocking performance of a SIP network?� The importance to a carrier is to understand how the addition of TRIP
will affect a SIP network. The addition of TRIP will add complexity and cost over a SIP network. This experiment will show the benefits.
University of Kansas19
TRIP Model�TRIP research questions to be addressed through
simulation.
�Model developed to evaluate each research issue.
�Model developed using Extend graphical modeling software.
�Model will simulate appropriate number of events (call requests) to provide a 99% confidence interval.
University of Kansas20
TRIP Model Description�TRIP Model Architecture:
� Two Location Server & Gateway Pairs� Each LS/GW pair has a dedicated call request generator� LS1 only sends calls to GW1 and LS2 only sends calls to GW2� GW1 has 48 DS-0 to destination prefix� GW2 has 24 DS-0 to destination prefix� Each LS is running TRIP� Each GW is running the TRIP-lite Client� TRIP-lite messaging is sent only when GW is at full trunk utilization.
This is the worst case-scenario. � Varying network topology via element-to-element propagation delay.
University of Kansas21
TRIP Model Description
PSTN/SS7
1 2 34 5 67 8 9
* 8 #
LS1 Call Generator
1 2 34 5 67 8 9
* 8 #
LS2 Call Generator
Lite
GW2
Lite
GW1
LS
LS1
LS
LS2
LS to LS CallRequest Rerouting
48 Trunks to PSTN
24 Trunks to PSTN
TRIP-lite UpdateMessaging
TRIP-lite UpdateMessaging
Generates TrafficLoad of 1% Call
Blocking
GeneratesVariableTraffic
Load which varies CallBlocking
LS-to-LS Delay LS-to-LS Delay
LS-to-GW Delay
LS-to-GW Delay
LS-to-GW Delay
LS-to-GW Delay
University of Kansas22
TRIP Model Description� Variable System
Parameters� LS-to-GW Propagation Delay� LS-to-LS Propagation Delay� Traffic Load� Gateway Trunk Failure
� System Performance Metrics� Overall System Call Blocking� LS Call Blocking� GW Call Blocking� Call Request Delivery Delay� Percentage Call Request
Rerouting between Location Servers
� Cumulative Blocked Calls during Gateway Trunk Failure
University of Kansas23
TRIP Simulation Results� System Parameters:
�Varied traffic loads ranging from 1% to 85% call blocking.
�LS-to-GW and LS-to-LS propagation delay was varied.�Delay values included:
�~0ms: elements co-located�24ms: cross country fiber connection�250ms: satellite link.
�GW trunk failure.
University of Kansas24
-33.86005 2468.962 4971.783 7474.605 9977.427 12480.25 14983.07 17485.89 19988.71 22491.53 24994.36 27497.18 30000-2.3315e-15
0.6717197
1.343439
2.015159
2.686879
3.358599
4.030318
4.702038
5.373758
6.045477
6.717197
7.388917
8.060637
8.732356
9.404076
10.0758
10.74752
11.41924
12.09095
12.76267
13.43439
14.10611
14.77783
15.44955
16.12127
16.79299
17.46471
18.13643
18.80815
Time
ValuePlotter, DE Multisim
Product 0 Product 1 Product 2 Product 3
TRIP Simulation Results� Impact of propagation delay and load variation on Call Blocking.� Results for system call blocking vs. time, 5% call blocking, LS-to-GW
delay variation, Blue: 0ms; Red: 24ms; Green: 250ms.
5%
Steady State tends to Erlang B Prediction of 5% Call Blocking
University of Kansas25
TRIP Simulation Results� System call blocking is driven by traffic load.� System call blocking follows Erlang B prediction.� Propagation delay and thus network topology does not impact system call
blocking performance.
System Call Blocking %LS to GW Delay
0102030405060708090
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Blo
ckin
g (%
)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250msErlang B
University of Kansas26
TRIP Simulation Results� For low LS-to-GW delay, LS call blocking follows Erlang B predictions.� For high delay (satellite link), blocking shifts from LS to GW.
� LS-to-LS delay variation does not cause blocking shift from LS to GW.
GW Call Blocking %LS to GW Delay
0
2
4
6
8
10
12
14
16
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)B
lock
ing
(%)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250ms
LS Call Blocking %LS to GW Delay
0102030405060708090
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Blo
ckin
g (%
)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250ms
University of Kansas27
TRIP Simulation Results� Impact of propagation delay and load variation on call request rerouting.� Call request rerouting is driven by traffic load.� The higher the traffic load the great percentage of reroutes.� Propagation delay and thus network topology does not impact the
percentage of call request reroutes.
Percent Call Request RerouteLS to GW Delay
0
5
10
15
20
25
30
35
40
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Rer
oute
% (%
)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250ms
Percent Call Request RerouteLS to LS Delay
0
5
10
15
20
25
30
35
40
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Rer
oute
% (%
)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250ms
University of Kansas28
TRIP Simulation Results� Impact of propagation delay and load variation on Call Request Delivery
to a GW.� LS-to-GW propagation delay will add directly to the call delivery delay.� The percentage of LS-to-LS delay added to total call delivery delay is
based on traffic load and call request rerouting.� Overall call delivery delay is the sum of both.
Call Request Delivery DelayLS to GW Delay
0
50
100
150
200
250
300
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Cal
l Req
uest
Del
iver
yD
elay
(ms)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250ms
Call Request Delivery DelayLS to LS Delay
0
20
40
60
80
100
120
140
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Cal
l Req
uest
Del
iver
yD
elay
(ms)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250ms
University of Kansas29
TRIP Simulation Results� The relationship between call request rerouting and LS-to-LS
delay can be illustrated by this predictive expression.
� At 24ms LS-to-LS propagation delay there are 4.3% calls rerouted and incur both the 24ms LS-to-LS delay plus the constant 4ms LS-to-GW delay.
� Subsequently, 95.7% of call requests do not get rerouted and only incur the 4ms LS-to-GW delay.
� Call Delivery Delay = 0.043*[24ms + 4ms] + 0.957*[4ms] = 5.032ms� The predicted value through simulation is 5.634ms. Thus the delivery
delay can be predicted given the simulated call request reroute percentage.
University of Kansas30
TRIP Simulation Results� Comparison of a TRIP-enabled Network to a SIP Network.� SIP call blocking is consistently higher than TRIP. � Given standard traffic assumptions Erlang B cannot be used to predict call
blocking in a SIP network.
System Call Blocking %LS to GW Delay
0102030405060708090
100
57.95 66.65 79.55 108.05 204.15 478.80
Traffic Intensity (Erlang)
Blo
ckin
g (%
)
0ms3.43ms6.86ms10.29ms13.72ms17.15ms20.58ms24.01ms125ms250msSIPErlang B
University of Kansas31
TRIP Model Demonstration� Demonstration of TRIP model simulating trunk failure on
GW1. 1% call blocking prior to failure.� At 50,000 seconds, GW1 suffers a T1 failure. This drops the
available trunks on GW1 from 48 to 24. Overall system resources drop from 72 DS-0 to 48.
� After failure, Erlang B predicts blocking of 22.1%.� At 125,000 seconds, restoral of failed trunk and 1% call
blocking. � At 200,000 seconds, simulation ends.
University of Kansas32
TRIP Simulation Results� Gateway trunk failure results� Cumulative number of blocked calls versus time.
0 12500 25000 37500 50000 62500 75000 87500 100000 112500 125000 137500 150000 162500 175000 187500 2000001
153.125
305.25
457.375
609.5
761.625
913.75
1065.875
1218
1370.125
1522.25
1674.375
1826.5
1978.625
2130.75
2282.875
2435
2587.125
2739.25
2891.375
3043.5
3195.625
3347.75
3499.875
3652
3804.125
3956.25
4108.375
4260.5
4412.625
4564.75
4716.875
4869
5021.125
5173.25
5325.375
5477.5
5629.625
5781.75
5933.875
6086
Time
ValuePlotter, DE Multisim
NumberExited 0 NumberExited 1 NumberExited 2 NumberExited 3
Failure
Restoral
time (seconds)
Blo
cked
Cal
ls
University of Kansas33
� TRIP results during trunk failure with varied propagation delay.
� LS-to-LS results also show propagation delay has no impact.
TRIP Simulation Results
Delay Location
Delay (ms)
Interval Average Slope (blocked/sec)
Calculated Call Blocking
(%)
Erlang B (%)
LS-to-GW 0ms Before Failure
0.00347 1.08% 1.0%
LS-to-GW 0ms After Failure 0.07013 21.8% 22.1%
LS-to-GW 0ms After Restoral 0.00378 1.2% 1.0%
LS-to-GW 24ms Before Failure
0.00331 1.03% 1.0%
LS-to-GW 24ms After Failure 0.07181 22.3% 22.1%
LS-to-GW 24ms After Restoral 0.00344 1.07% 1.0%
LS-to-GW 250ms Before Failure
0.00352 1.09% 1.0%
LS-to-GW 250ms After Failure 0.07217 22.4% 22.1%
LS-to-GW 250ms After Restoral 0.00327 1.02% 1.0%
University of Kansas34
TRIP Simulation Results• System reaction to a state change is dependent on traffic
load.• As the traffic load is increased, the system reaction time
to the state change will decrease.• For a 1% call blocking, the TRIP system will react to the
state change in approximately 3.1-9.3 second.• 15% in 2.3-6.9 second & 85% in 0.38-1.14 second.• Propagation delay during a failure scenario does not
impact the system reaction to new state.
University of Kansas35
TRIP Conclusions� Network topology does not impact system blocking
probability. In a TRIP-enabled network, the system blocking will be driven by traffic load.� This result impacts geographic deployment of location servers to
support the network. From a system blocking standpoint, designers do not need to be concerned with propagation delay but must be concerned with traffic load.
� Overall system blocking will follow Erlang B given specified values for traffic load and call holding time. � This result allows designers to implement a correctly sized TRIP
network based on forecasted customer usage. This would impact number of trunks to support a given destination prefix, number of gateways in a geographic area, and number location servers in the network.
University of Kansas36
TRIP Conclusions� As LS-to-GW delay is increased towards a satellite link delay
(250ms), loss of knowledge about the current state of the system causes call blocking to increase at the GW. � This result places a limit on implementation options. TRIP messaging
can incur propagation delay equivalent to cross country fiber links but satellite links should not be considered.
� Propagation delay, LS-to-GW and LS-to-LS, does not impact the percentage of reroutes in the system. The traffic intensityis the driving factor. � This result dictates that designers be concerned with traffic load and
not propagation delay when addressing TRIP rerouting functionality.
University of Kansas37
TRIP Conclusions� LS-to-GW propagation delay will add directly to the call
delivery delay. For LS-to-LS delay only a percentage of the propagation delay will add into the total call delivery delay. And that amount will be dependent upon the traffic load. � This issue impacts the delay budget. The result indicates that any delay
between the LS and GW must be added to overall call setup delay. While, only a percentage of the delay between LS and LS should be added.
� SIP blocking is consistently higher than TRIP and higher than what would be predicted by Erlang B.� TRIP provides a SIP network with lower blocking. It benefits the carrier with
less provisioning, gateway dynamic resource information available at the proxy, optimum path routing, and also better blocking performance.
University of Kansas38
TRIP Conclusions� The time required for a TRIP system to react to a change in
state (i.e., gateway trunk failure) is based on traffic load. As the traffic load is increased, the system reaction time to the state change will decrease. Additionally, the results show that propagation delay during a failure scenario does not impact the system reaction to new state.� When a failure happens the TRIP network will react within a
reasonable time interval and tend toward the new steady state.
University of Kansas39
Next Steps� Simulation of TRIP-lite network with added update
messaging. � Simulation of TRIP network synchronization (TRIP Routing
Convergence).� Lab evaluation of Vendor TRIP-lite equipment and software.� Lab evaluation of Vendor Interior Administrative Domain
Routing (I-TRIP) equipment and software.� Lab evaluation of Vendor Exterior Administrative Domain
Routing (E-TRIP) equipment and software.� Lab evaluation of a TRIP network with all TRIP entities.
University of Kansas40
Acknowledgements
�Special Thanks to:�Dr. Victor Frost�Dr. Joseph Evans�Dr. Gary Minden�The Cisco TRIP Marketing and Development
Team�And Very Special Thanks to Sprint, Network
Services Executive Management.
University of Kansas41
Questions