Transcript Planning of Multiservice IP Networks

s
IP Network Engineering Challenges
Dr. Thomas Bauschert
Senior Consultant
Network Planning and Design
Siemens AG, München
Email: [email protected]
IP Network Engineering Challenges
T. Bauschert
11.05.2001
1
s
Outline
• Network Architecture
• IP Traffic Specification
• IP Network Dimensioning
• IP Traffic Engineering and QoS Provisioning
• Further important Engineering Issues
• Special Topic: MPLS - DiffServ: Combination of Traffic Engineering and QoS Provisioning
IP Network Engineering Challenges
T. Bauschert
11.05.2001
2
s
Network Architecture
IP Network Engineering Challenges
T. Bauschert
11.05.2001
3
s
Network Architecture
• State-of-the-art network architecture:
• switched (MPLS) IP core network (highly meshed)
• multiple homed distribution/access routers
Internet
Exchange
Interconnect
Router
Partially meshed
STM-64 / 160 
GigaEth
Core
Router(320G)
Ring
STM-16/64 / 32 *
Distribution
Router(40G)
Data Center
Router
Ring
STM-4, GigaEth / 32 *
Access Router
Aggr. Devices
Data
Centers
STM-1/4,
GigaEth
STM-1
Access
Devices
GigaEth
MGW
100bT
E1, E3, STM-1 LL
Local
Exchange
IP Network Engineering Challenges
T. Bauschert
11.05.2001
4
s
Network Architecture
• Future challenges:
• scaleable/reliable network architecture (to accommodate huge IP traffic
growth): Tbit/s-routers required in near future
• switched (MPS) optical core network (ASON): interaction of IP and
optical layer (similar to IPoATM)
state-of-the-art
future
ASON
ATM
virtual meshing via ATM PVCs
physical meshing
virtual meshing via wavelenghts
router throughput < 1 Gbit/s,
STM-16 I/F
router throughput < 1 Tbit/s,
STM-64 I/F
router throughput < 1 Tbit/s
n x STM-64 DWDM I/F
IP Network Engineering Challenges
T. Bauschert
11.05.2001
5
s
IP Traffic Specification
IP Network Engineering Challenges
T. Bauschert
11.05.2001
6
s
IP Traffic Specification (for Network Planning Purposes)
• State-of-the-art:
• traffic model stream/elastic traffic:
stream traffic parameters: offered traffic + effective bitrate
elastic traffic parameters: flow arrival rate + average flow size
• traffic matrix generation methodology
Voice
Video
Traffic Class 1
Traffic Class 2
Stream
Traffic
QoS 1
•
•
•
•
•
•
WWW
FTP
Traffic Class 3
Stream
Traffic
QoS 2
Elastic
Traffic
QoS 1
•
arrival rate + duration
or: offered traffic
mean + peak bitrate
blocking probability
connection setup delay
packet delay / jitter / loss
•
•
•
•
•
...
Services/
Applications
...
Traffic
Classes
Elastic
Traffic
QoS 2
file arrival rate +
average file size
offered traffic volume
average throughput
or: average transfer time
for file of specific size
delay before data transfer
packet delay / jitter / loss
IP Traffic
Types
(w/w.o. QoS
guarantee)
Traffic Parameter
QoS Param. (flow)
QoS Param. (packet)
Server
Site
2)
• Challenges:
• traffic classification: how many classes are really necessary?
• which QoS metrics should be applied? (e.g. blocking probability for
stream traffic is only reasonable in case of CAC)
• point-to-multipoint traffic description
IP Network Engineering Challenges
5)
6)
1)
User
Site
ISPs /
Internet
4)
IP Network
Cache
3)
strong relationship
to QoS mechanisms
T. Bauschert
11.05.2001
7
s
IP Network Dimensioning
IP Network Engineering Challenges
T. Bauschert
11.05.2001
8
s
IP Network Dimensioning
• State-of-the-art:
• single link dimensioning with multirate Erlang-B
(stream) and M/G/R-PS (elastic, ideal TCP behavior)
model
• dimensioning for tree-type access networks
• separate dimensioning for elastic / stream traffic portions
average time
to transfer a file
of length x
file length
E{T ( x)} 
ratio C / rpeak
x  E 2 ( R, R ) 
1 

rpeak 
R(1   ) 
utilization:  = (·xmean)/C
peak bit rate
blocking
blocking
probability
probabilityofof
traffic
trafficclass
classss
solution
solutionofofthe
the
characteristic
characteristic
polynom
polynom
effective
effectivebitrate
bitrate
ofoftraffic
trafficclass
classss
service discrimination
(realized by link buffer scheduler)
average TCP
throughput
offered
elastic
traffic
volume
elastic traffic
offered
stream
traffic
stream traffic
blocked
stream
traffic
carried
stream
traffic
admission control
IP Network Engineering Challenges
Bs 
1   rs (C )
M C
 E1  , 
C
 d d
1
M
bitrate
bitrateofofthe
the
equivalent
equivalentsingle
single
server
serversystem
system
total
totaloffered
offered
traffic
trafficvolume
volume
link capacity
T. Bauschert
11.05.2001
9
s
IP Network Dimensioning
• Challenges:
• link dimensioning model improvements:
- dimensioning formula for short flows
- M/G/R-PS extension for multiple rpeak
- consideration of QoS mechanisms and multiple QoS levels
• network dimensioning algorithm (similar to the well-known unified algorithm for PSTN and ATM networks)
with following features:
- integrated (IGP) routing optimization
- consideration of constraints imposed by TE and QoS mechanisms like MPLS, OMP, DiffServ
- multiple load period dimensioning - point-to-any dimensioning (for DiffServ networks)
- consideration of restoration capabilities (e.g. via MPLS)
- dimensioning for multiple QoS metrics
• integration of dimensioning algorithm and TE system in automated planning and engineering system
IP Network Engineering Challenges
T. Bauschert
11.05.2001
10
s
IP Traffic Engineering
and QoS Provisioning
IP Network Engineering Challenges
T. Bauschert
11.05.2001
11
s
IP Traffic Engineering and QoS Provisioning
• State-of-the-art:
• separate application of Traffic Engineering and QoS mechanisms (multipath routing (OMP), MPLS LSP
adaption, DiffServ, IntServ) in IP networks
• Challenges:
• performance evaluation of TE / QoS mechanisms
• use of TE for fast load adaption and restoration
• development of optimum TE control algorithm (objectives/constraints of TE?)
• information exchange/interaction of TE mechanism and routing
• combination of different mechanisms: MPLS-OMP, MPLS-DS
- requires protocol enhancements > new IETF drafts
- driver: search for optimum tradeoff between
overprovisioning and complexity
- strong relationship to SLA formulation
- will admission control be really neccessary?
• end-to-end QoS provisioning: reasonable scenarios
IP Network Engineering Challenges
Multipath
Routing (OMP)
MPLS TE
DiffServ
T. Bauschert
11.05.2001
12
s
Further important
Engineering Issues
IP Network Engineering Challenges
T. Bauschert
11.05.2001
13
s
Further important Engineering Issues
• Optimization tasks concerning routing protocols:
• IGP (OSPF, ISIS) design / optimization rules
• EGP (BGP-4) design / optimization rules (e.g. application and # of route reflectors, confederations etc.)
• traffic induced by routing protocol
• performance evaluation of routing protocols
• Optimization of Data Center (server site) locations
• Engineering of Data Centers
IP Network Engineering Challenges
T. Bauschert
11.05.2001
14
s
Special Topic:
MPLS-DiffServ*
*partially taken from MPLS2000 Conference material
IP Network Engineering Challenges
T. Bauschert
11.05.2001
15
s
MPLS-DiffServ
IP Network Engineering Challenges
T. Bauschert
11.05.2001
16
s
MPLS-DiffServ
IP Network Engineering Challenges
T. Bauschert
11.05.2001
17
s
MPLS-DiffServ
IP Network Engineering Challenges
T. Bauschert
11.05.2001
18
s
MPLS-DiffServ
Overview
OMP
MPLS-OMP
increasing
complexity!
MPLS
DiffServ (DS)
IntServ
DS over MPLS
(or: MPLS - DS):
•E-LSP
•L-LSP
•TE not CoS aware!
DS aware MPLS TE:
•TE is CoS aware
DS aware MPLS TE + RSVP
CAC for rt-Traffic (e.g. voice)
IP Network Engineering Challenges
reinvention of ATM!
T. Bauschert
11.05.2001
19
s
MPLS-DiffServ
IETF Drafts
MPLS TE:
• “Requirements for Traffic Engineering Over MPLS”
RFC2702, Informational RFC
• “RSVP-TE: Extensions to RSVP for LSP Tunnels”
draft-ietf-mpls-rsvp-lsp-tunnel-07.txt, Aug 2000
• “Constraint-Based LSP Setup using LDP”
draft-ietf-mpls-cr-ldp-04.txt, Jul 2000
• “Extensions to ISIS for TE”
draft-ietf-isis-traffic-03.txt, Sept 2000
• “TE extensions to OSPF”
draft-katz-yeung-ospf-traffic-03.txt, Oct 2000
DS over MPLS:
• “MPLS Support of Diff-Serv”
draft-ietf-mpls-diff-ext-07.txt, Aug 2000
DS aware MPLS TE:
• “Requirements for support of Diff-Serv-aware MPLS
Traffic Engineering”
draft-lefaucheur-diff-te-reqts-00.txt, Jul 2000
• “Extensions to IS-IS, OSPF, RSVP and CR-LDP for
support of Diff-Serv-aware MPLS TE”
draft-lefaucheur-diff-te-ext-00.txt, Jul 2000
IP Network Engineering Challenges
T. Bauschert
11.05.2001
20
s
MPLS-DiffServ
Example: MPLS TE
IP Network Engineering Challenges
T. Bauschert
11.05.2001
21
s
MPLS-DiffServ
Example: DS over MPLS
IP Network Engineering Challenges
T. Bauschert
11.05.2001
22
s
MPLS-DiffServ
Example: DS aware MPLS TE
IP Network Engineering Challenges
T. Bauschert
11.05.2001
23
s
MPLS-DiffServ
Example: DS aware MPLS TE
IP Network Engineering Challenges
T. Bauschert
11.05.2001
24
s
MPLS-DiffServ
Example: VoMPLS - DS aware MPLS TE with RSVP CAC
IP Network Engineering Challenges
T. Bauschert
11.05.2001
25
s
MPLS-DiffServ
VoMPLS: DS aware MPLS TE with RSVP CAC
n “ultimate QoS” solution for VoMPLS:
n QoS never degrades
n automatic/dynamic traffic engineering of voice
(exceeds today’s TDM TE capabilities)
n excess calls get rejected if/when EF-capacity exceeded
n traffic patterns do not have to be known before
n This level of sophistication is only useful in some
environments
n Under construction at IETF
IP Network Engineering Challenges
T. Bauschert
11.05.2001
26
s
MPLS-DiffServ
IP Network Engineering Challenges
T. Bauschert
11.05.2001
27
s
Additional
Slides
IP Network Engineering Challenges
T. Bauschert
11.05.2001
28
s
MPLS-DiffServ
IP VPN-Concepts: HOSE Model
IP Network Engineering Challenges
T. Bauschert
11.05.2001
29
s
MPLS-DiffServ
IP VPN Concepts: PIPE Model
IP Network Engineering Challenges
T. Bauschert
11.05.2001
30
s
MPLS-DiffServ
Necessary Enhancements for DS aware MPLS TE
Current IGP extensions for TE:
n advertise “unreserved TE bandwidth” (at each preemption level)
Proposed IGP extensions for DS aware TE:
n Class-Types= group of Diff-Serv classes sharing the same bandwidth constraint (eg
AF1x and AF2x)
n advertise “unreserved TE bandwidth” (at each preemption level) for each Class-Type
Current LSP-signalling extensions for TE:
n at LSP establishment signal TE tunnel parameters (label, explicit route, affinity ,
preemption,…)
Proposed LSP-signalling extensions for DS aware TE:
n also signal the Class-Type
n perform Class-Type aware CAC
Current Constraint Based Routing for TE:
n compute a path such that on every link there is sufficient “unreserved TE
bandwidth”
Proposed Constraint Based Routing for DS aware TE:
n same CBR algorithm but satisfy bandwidth constraint over the “unreserved
bandwidth for the relevant Class-Type” (instead of aggregate TE bandwidth)
IP Network Engineering Challenges
T. Bauschert
11.05.2001
31