diana: scenarios for qos based integration of ip and atm
DESCRIPTION
Presentation of research project at the EU IP/ATM cluster meeting in February 1999. EU IP/ATM cluster meeting, Lausanne Switzerland February 11, 1999TRANSCRIPT
DIANA: Scenarios for QoS based DIANA: Scenarios for QoS based integration of IP and ATMintegration of IP and ATM
EU IP/ATM cluster meeting, Lausanne Switzerland Feb 11, 1999
John Loughney
Nokia Research Center
MotivationMotivation
ATM and IP evolution have some commonalties: ATM and IP evolution have some commonalties:
need for QoS ,need for QoS ,
need for more capacity in the network,need for more capacity in the network,
need for efficient resource usage (multicasting, multiplexing, need for efficient resource usage (multicasting, multiplexing, dynamism etc..),dynamism etc..),
ATM is an established transport layer technology, but IP dominatATM is an established transport layer technology, but IP dominates in es in application side & is a hot technologyapplication side & is a hot technology
in order to evaluate the options we should look into areas wherein order to evaluate the options we should look into areas where ATM ATM and IP overlap, conflict and complete each other: and IP overlap, conflict and complete each other: QoS based QoS based interworking between ATM and IPinterworking between ATM and IP
Objectives of the projectObjectives of the projectTo develop, integrate, validate and demonstrate resource reservaTo develop, integrate, validate and demonstrate resource reservation tion and traffic control functionalities which seamlessly interoperatand traffic control functionalities which seamlessly interoperate e between ATM and IP networks to achieve QoS. between ATM and IP networks to achieve QoS.
At the boundary between the ATM and IP domains, DIANA will At the boundary between the ATM and IP domains, DIANA will translate between IP QoS and ATM QoS mechanisms. Alternative IP translate between IP QoS and ATM QoS mechanisms. Alternative IP QoS schemes are addressed (RSVP, SPR, SIMA/diff. serv.) and theiQoS schemes are addressed (RSVP, SPR, SIMA/diff. serv.) and their r relationship to ATM QoS is studied. relationship to ATM QoS is studied.
The experimental prototyping in DIANA will allow the investigatiThe experimental prototyping in DIANA will allow the investigation, on, testing and validation of different approaches for the convergentesting and validation of different approaches for the convergence of ce of IP and ATMIP and ATM
Project PartnersProject Partners
Association Swisscom & Ascom Association Swisscom & Ascom SwitzerlandSwitzerlandUniversity of Stuttgart University of Stuttgart GermanyGermanyTelenor Research & DevelopmentTelenor Research & Development NorwayNorwayFinsiel Finsiel ItalyItalyFlextelFlextel ItalyItalyNokia Nokia FinlandFinlandAscom Hasler AGAscom Hasler AG SwitzerlandSwitzerlandEcole Polytechnique Federale de LausanneEcole Polytechnique Federale de Lausanne SwitzerlandSwitzerlandSwisscomSwisscom SwitzerlandSwitzerlandTelscom AGTelscom AG SwitzerlandSwitzerland
QoS strategies for IPQoS strategies for IP
Resource Reservation Protocol (RSVP)Resource Reservation Protocol (RSVP)per flow reservation per flow reservation
role: in access networks, for real time applicationsrole: in access networks, for real time applications
Scalable Reservation Protocol (SRP)Scalable Reservation Protocol (SRP)adaptive, aggregated reservationsadaptive, aggregated reservations
role: core and access, for adaptive applicationsrole: core and access, for adaptive applications
Differentiated Services/SIMADifferentiated Services/SIMAaggregating, priority basedaggregating, priority based
perfect match for core networksperfect match for core networks
Resource Reservation Protocol (RSVP)Resource Reservation Protocol (RSVP)designed for an Integrated Service Internet designed for an Integrated Service Internet (Guaranteed & Controlled Load)(Guaranteed & Controlled Load),,
provides endprovides end--toto--end QoS reservations for uniend QoS reservations for uni--directional flows,directional flows,
operates on top of IP4 and IP6, out band "signaling"operates on top of IP4 and IP6, out band "signaling"
supports multicasting, supports multicasting,
receiver oriented, relies on periodical refreshments (soft statereceiver oriented, relies on periodical refreshments (soft state))
scalability problems with large numbers of flowsscalability problems with large numbers of flows
S1
R1
R2
Path
Resv
Multicast Packet FlowPath message (multicast)Resv message (unicast)
PROPROAggregation of traffic Aggregation of traffic streams at the ingress to streams at the ingress to ATM: + ScalabilityATM: + Scalability
Traffic descriptor & QoS Traffic descriptor & QoS parameter based parameter based resource reservation resource reservation endend--toto--end: Exact & end: Exact & reliable guaranteereliable guarantee
RSVP over ATM FrameworkRSVP over ATM Framework
CONCONLayered routing, Layered routing, addressing & traffic addressing & traffic controlcontrol
Different signaling Different signaling protocolsprotocols
RSVP over ATM ImplementationRSVP over ATM Implementation
CLIP address resolution: Proof of CLIP address resolution: Proof of conceptconcept
UNI signaling with dynamic reUNI signaling with dynamic re--negotiationnegotiation
However, focus is on traffic However, focus is on traffic controlcontrol
FlowFlow--toto--VC aggregationVC aggregation
Dynamic, threshold based Dynamic, threshold based VC bandwidth renegotiation VC bandwidth renegotiation (cf. ACTS REFORM) (cf. ACTS REFORM)
Interworking ScenariosInterworking Scenarios
Overlay model (Scenario 1)Overlay model (Scenario 1) Peer model (Scenario 2)Peer model (Scenario 2)
Common transport layer
Integration unit
Technology X Technology Y
• what is the common layer? RTP, UDP (TCP), IP?
• Majority of applications use IP
• Mapping of addressing, QoS parameters, control plane interworking
Technology X Technology Y
Transport Layer A
• translation of user plane, find common "application" layer: RTP ... Results into application specificinterworking
•translation of control plane
• translation of addressing, QoS parameters, ...
Integration unit Transport Layer B
RSVP over ATMRSVP over ATM
Overlay model where common layer is IP and ATM as a link layer, Overlay model where common layer is IP and ATM as a link layer, control plane mappings required, IETF RFCscontrol plane mappings required, IETF RFCs
RSVP peering with ATMRSVP peering with ATMIntegration Unit terminates both RSVP and ATM signaling.Two approaches for setting up the ATM section for originating ATM host: 1) it is set up only after RSVP section is done 2) sequentiallyIn some case loss of data may occur, QoS negotiation problematic,application level interworking on user plane.
RSVP & DiffServ Interworking Over ATMRSVP & DiffServ Interworking Over ATM
Diff. Serv over ATM
Sending host
Receiving hostStub network Stub networkTransit networkER1 ER2
Int-serv Int-servDiff-serv
Data stream: best effortRSVP Path
RSVP ResvRSVP Resv
DACSRSVP Resv
Data stream: QoS
Note: diff-serv bits can be set at ER1:
Data stream Data stream with diff-serv bits Data stream with diff-serv bits
Dynamic Bandwidth ManagementDynamic Bandwidth Management
RSVP over ATM IssuesRSVP over ATM Issues
Management of QoS data, VCs and VC for control Management of QoS data, VCs and VC for control messaging (RSVPmessaging (RSVP--messaging)messaging)
ITU SG11 looking into possibility to embed RSVP ITU SG11 looking into possibility to embed RSVP messaging into Generic Identifier IE in ATM signaling messaging into Generic Identifier IE in ATM signaling
change of QoS: ATM allows change of PCR and SCR, not change of QoS: ATM allows change of PCR and SCR, not service category, establishment of new VCservice category, establishment of new VC
release of VCrelease of VC
set up of a VC is a costly operation; implies flow to VC set up of a VC is a costly operation; implies flow to VC aggregation among RSVP sessionsaggregation among RSVP sessions
multicast supportmulticast support
Scalable Reservation Protocol (SRP)Scalable Reservation Protocol (SRP)Inband signaling for "Controlled Load" servicesInband signaling for "Controlled Load" services
routers aggregate the flows on output portsrouters aggregate the flows on output ports
sources request resources; routers estimate reserved bandwidth tsources request resources; routers estimate reserved bandwidth to o decide on admission and propagate requests; receivers send feedbdecide on admission and propagate requests; receivers send feedback ack to sourceto source
sources adjust traffic based on the feedback from receivers sources adjust traffic based on the feedback from receivers
failed reservation requests are degraded (to best effort)failed reservation requests are degraded (to best effort)
Feedback
Sender Data & reservations Receiver
Router
SRP Reservation MechanismSRP Reservation MechanismDESIGN GOALSDESIGN GOALS
Scalability: No SRP Scalability: No SRP router manages perrouter manages per--flow flow statestate
Absolute reservationAbsolute reservation
Soft state: Flow of data Soft state: Flow of data packets marked as packets marked as reserved or request reserved or request maintain or upgrade the maintain or upgrade the reservationreservation
No traffic descriptor and No traffic descriptor and QoS parameter based QoS parameter based (explicit) signaling(explicit) signaling
IP layer solutionIP layer solution
SRP Implementation FrameworkSRP Implementation FrameworkPROPRO
Realizes absolute Realizes absolute reservation with a reservation with a scaleable architecturescaleable architecture
Can be embedded in IP Can be embedded in IP DiffServ architectureDiffServ architecture
Relatively simple traffic Relatively simple traffic control in routerscontrol in routers
Builds upon the LINUX Builds upon the LINUX DiffServ architecture DiffServ architecture that DIANA (EPFL) has that DIANA (EPFL) has coco--developeddeveloped
CONCON
No explicit QoS No explicit QoS parametersparameters
Feedback controlFeedback control
Sensitive to route Sensitive to route changeschanges
Policing more complexPolicing more complex
SRP over ATMSRP over ATM
Time
BW
ATM signaling latency
= ATM reservationbased on estimates
SRP Performance ExampleSRP Performance ExampleObserved performanceObserved performance
SRP builds up reservation SRP builds up reservation gradually (starting from BE)gradually (starting from BE)
Transient phasesTransient phases
SRP provides stable SRP provides stable reservationreservation
Further workFurther work
Completion of the LINUX Completion of the LINUX framework: IP over ATM, framework: IP over ATM, DiffServ & SRP queuing DiffServ & SRP queuing disciplinediscipline
PolicingPolicingTime
0e+0 5e+4 1e+5 2e+5 2e+5
Res
erve
d R
ate
0,0
0,1
0,2
0,3
0,4
0,5
0,6
2 SRP sources(long ON and OFF periods)
SRP over ATM IssuesSRP over ATM IssuesDue to the inband nature of the SRP, overlay model is the Due to the inband nature of the SRP, overlay model is the most naturalmost natural
SRP builds up reservations gradually, while ATM expects SRP builds up reservations gradually, while ATM expects the connection parameters to be known at connection set the connection parameters to be known at connection set up. up.
A request packet that is destined to an ATM VC without A request packet that is destined to an ATM VC without enough capacity can cause:enough capacity can cause:
a new connection set up a new connection set up
rere--negotiation of an existing connectionnegotiation of an existing connection
These are costly operations, anticipation of future These are costly operations, anticipation of future connections neededconnections needed
SSimple imple IIntegrated ntegrated MMedia edia AAccessccessWHY DIFFERENTIATED SERVICES?WHY DIFFERENTIATED SERVICES?
Integrated Services provide absolute Integrated Services provide absolute & reliable QoS& reliable QoS
BUT perBUT per--flow state has to be flow state has to be established and managed: Not established and managed: Not scaleablescaleable
DS achieve scalability by classifying DS achieve scalability by classifying and marking packets at the ingress and marking packets at the ingress to a DS networkto a DS network
Profile based accessProfile based access
DS can provide relative QoS using DS can provide relative QoS using simple, scaleable mechanisms simple, scaleable mechanisms
HOW DOES SIMA IMPLEMENTHOW DOES SIMA IMPLEMENT
DIFFERENTIATED SERVICES?DIFFERENTIATED SERVICES?
Customer profile expressed in Customer profile expressed in terms of a nominal bit rateterms of a nominal bit rate
RealReal--time or non realtime or non real--time time service selection for a flowservice selection for a flow
Drop priority of a flow Drop priority of a flow dynamically depends on dynamically depends on current bit rate to NBR ratiocurrent bit rate to NBR ratio
8 drop preference and 2 delay 8 drop preference and 2 delay priority (rt/nrt) levelspriority (rt/nrt) levels
Defines access & core router Defines access & core router functionalityfunctionality
SIMA / DiffServSIMA / DiffServ
• Access node (A):• knows NBR and measures actual bit rate of flow (MBR)
• determines and sets the drop preference of each packet (from 0 to 6)
• Core network node (C):• discards packets based only on the drop preference of the packet and buffer occupancy levels
• places the accepted packet either in the real-time or nrt buffer
• does not need to know anything about the NBR or the traffic process of separate connections
• no capacity reservation (CAC or RSVP)
CE = Customer EquipmentA = Access NodeC = Core Node
CE A
C
C
C
ASIMA network CE
C
SIMA Drop Preference LevelsSIMA Drop Preference Levels
Moderate quality: usually small packet loss ratio except for busy hours
Good quality: Small packet loss even during busy hours
High quality, packet losses only during exceptional traffic peaks
Excellent quality
Reserved for non-SIMA services with resource reservation
Satisfactory quality: from time to time very high packet loss ratio
Suitable for best effort during busy hours
Unusable during busy hours: suitable for BE during idle hours
DP =7
Drop Preference
DP=6
DP =5
DP =4
DP =3
DP =2
DP =1
DP =0
NBR/4
NBR/2
NBR
2 NBR
4 NBR
8 NBR
16 NBR
Actual Bit Rate
If the network is dimensioned properly:
Packet scheduling and BufferingPacket scheduling and Buffering
DP = preference of the incoming packetDPa = allowed drop preference level
Mrt = occupancy level of real-time bufferMnrt = occupancy level of nrt buffer
DPa = f(Mrt, Mnrt)
DP ≥ DPa rt/nrt
+
nrt
rtMrt
Mnrt
no (discard)
yes
SIMA Implementation FrameworkSIMA Implementation FrameworkSIMA in the DIANA IUSIMA in the DIANA IUAccess node functionalityAccess node functionality
Measure & classify data into Measure & classify data into flowsflows
Check against NBRCheck against NBR
Calculate priorityCalculate priority
Core node functionalityCore node functionality
Drop & delay priority based Drop & delay priority based on DS bits and buffer on DS bits and buffer occupancyoccupancy
OptionsOptions
Dynamic NBR (RSVP & Dynamic NBR (RSVP & DiffServ)DiffServ)
Exploiting ATMExploiting ATM´́s flexibility s flexibility as L2as L2
SIMA Performance ExampleSIMA Performance Example
Example exhibits weight (NBR) Example exhibits weight (NBR) proportional fairnessproportional fairness
Further research required to Further research required to investigateinvestigate
the impact of dynamic the impact of dynamic priority assignment during priority assignment during congestion epochscongestion epochs
the interaction with TCPthe interaction with TCP´́s s flow and congestion flow and congestion controlcontrol
the behavior in large scale the behavior in large scale networksnetworks
Nominal Bit Rate
0.0 0.1 0.2 0.3
Nor
mal
ised
Thr
ough
put
0.0
0.1
0.2
0.3
0.4
0.5
0.6
1 Node, 5 TCP like sources
SIMA is developed for IP, could be reworked for ATM (in SIMA is developed for IP, could be reworked for ATM (in theory)theory)
Uses ATM as a link between SIMA routersUses ATM as a link between SIMA routers
Usual IP over ATM difficultiesUsual IP over ATM difficulties
SIMA Over ATM IssuesSIMA Over ATM Issues
Comparison of IP QoS strategies & ATM QoSComparison of IP QoS strategies & ATM QoS
RSVP SRP DiffServ UNI4.0
Receiver requests reservation inresponse to a PATH message, uni-
directional
Sender requests and maintains areservation by marking packets
Service Level Agreement, betweenuser and the ISP.
Sender sets up a bi-directional VC
Default best-effort service Default best-effort service Default best-effort No connectivity if set-up fails
Heterogeneous QoS within amulticast session
Multicast under study(Homogeneous QoS planned)
Multicast (still under study) Point-to-multipoint VCs with ahomogeneous QoS
Dynamic QoS: RESV can alter thereservation at any time
Dynamic QoS: Request is signalledimplicitly and dynamically by
marking packets
No means for signalling defined,static
Static QoS, negotiated at set-up(Q.2963.x now specifies sender
controlled modificationprocedures)
Multiple reservation filter styles toselect different senders in a
multipoint-to-multipoint scenario
n.a. Behaviour Aggregate classifier,Multi Field classifier
Point-to-multipoint
Soft state: Messages are resentperiodically
Soft, aggregated state in routers n.a. Hard state: Connections have to beexplicitly released
Guaranteed, controlled load andbest-effort service
Service similar to controlled load Many services (e.g. relativeguarantees and absolute)
CBR, rt/nrt-VBR, ABR, (GFR),UBR
SummarySummary
•Nominal bit rate•Relative priority
•Dynamic priority dependent on current bit
rate to NBR ratio
RSVP over ATM SRP SIMA
•Layered architecture •IP architecture •IP DiffServ
•Traffic & QoS parameter based
•Absolute bandwidth reservation without per-
flow state
•Most complex traffic control architecture •Control loop
Trial ScenarioTrial Scenario
More realistic traffic More realistic traffic scenarioscenario
RSVP over ATMRSVP over ATMSignaling (reSignaling (re--negotiation) negotiation) becomes more expensivebecomes more expensive
SRPSRPDelay in control loopDelay in control loopPolicing trialPolicing trial
SIMASIMAInteraction with TCP flow & Interaction with TCP flow & congestion controlcongestion control
DISTRIBUTED INFRASTRUCTUREDISTRIBUTED INFRASTRUCTURE
Planned Performance ExperimentsPlanned Performance ExperimentsRSVP over ATMRSVP over ATM
How does aggregation affect How does aggregation affect individual QoS?individual QoS?
Dynamic BW reDynamic BW re--negotiation: negotiation: TradeTrade--off between reservation off between reservation efficiency and signaling loadefficiency and signaling load
SRPSRP
Dynamic behavior in transient Dynamic behavior in transient phasesphases
PolicingPolicing
SIMASIMA
Impact of dynamic markingImpact of dynamic marking
Interaction with TCPInteraction with TCP
Application controlled Application controlled renegotiationrenegotiation
Exploits RSVPExploits RSVP´́s dynamic s dynamic rere--negotiation capabilitiesnegotiation capabilities
RVBR service for traffic RVBR service for traffic descriptor calculationdescriptor calculation
Reference ApplicationsReference Applications
VATVAT
ARMIDA VoD application ARMIDA VoD application (MPEG4): Adapts to (MPEG4): Adapts to available network available network resourcesresources
Synthetic sourcesSynthetic sources
Design of the Integration UnitDesign of the Integration UnitHW based on FlextelHW based on Flextel´́s Multi Purpose Switch/Routers Multi Purpose Switch/Router
open SW environment (Linux)open SW environment (Linux)
started with RSVP over ATM scenariostarted with RSVP over ATM scenario
Design of the Integration UnitDesign of the Integration Unit
HW based on FlextelHW based on Flextel´́s Multi Purpose Switch/Routers Multi Purpose Switch/Routeropen SW environment (Linux)open SW environment (Linux)
started with RSVP over ATM scenariostarted with RSVP over ATM scenario
ConclusionConclusion
DIANA platform has the ability to verify a number of approachesDIANA platform has the ability to verify a number of approaches
the project aims to provide answers to evolution considerationsthe project aims to provide answers to evolution considerations
test and evaluate the scenarios under considerationtest and evaluate the scenarios under consideration
year 1 started with RSVP over ATM scenarioyear 1 started with RSVP over ATM scenarioRSVPRSVP--ATM peer model seems to be problematic: no real application ATM peer model seems to be problematic: no real application need, no immediate market needneed, no immediate market need
year 2 starts:year 2 starts:proof of RSVP over ATM scenarioproof of RSVP over ATM scenario
SRP over ATM promising new approach SRP over ATM promising new approach
SIMA/Diff. Serv. seems more relevant than peering RSVP with ATM SIMA/Diff. Serv. seems more relevant than peering RSVP with ATM