Lecture 9 Quality of Service in ad hoc wireless networks Yevgeni Koucheryavy Department of Communications Engineering Tampere University of Technology yk@cs.tut.fi Lectured by Jakub Jakubiak
QoS statement 1/2 Quality of Service (QoS) The performance level of service offered by the network to the user The goal of QoS frameworks To achieve a more deterministic network behavior Information carried by the network can be better delivered and network resources can be better utilized A network can offer different kinds of services to the users A service can be characterized by a set of measurable pre-specified service requirements such as Minimum bandwidth, maximum delay, maximum delay variance (jitter), maximum packet loss rate After accepting a service request from the user The network has to ensure that the service requirements of the user s flow are met throughout the duration of the flow Problem with ad hoc networks Hard to obtain this information Links constantly changing: node mobility, environmental affects, etc. 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 2
QoS statement 2/2 Hard QoS Guarantee parameters such as delay, jitter, bandwidth Required for mission-critical applications E.g., air traffic control, nuclear reactor control Not feasible in MANETs Soft QoS Aim to meet QoS goals Loss in QoS degrades application but does not have disastrous consequences E.g., voice, video Most research focuses on providing soft QoS QoS parameters Bandwidth Delay Delay jitter Security Network availability Battery life 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 3
Why QoS is hard in MANETs? 1/2 MANETs have certain unique features Dynamic network topology Flow stops receiving QoS provisions due to path break New paths must be established, causing data loss and delays Imprecise state information Link-specific state information Such as bandwidth, delay, jitter, loss rate, stability, cost, distance, power Subject to continuous changes Flow-specific state information Session ID, source address, destination address, max/min bandwidth requirements, max delay, max jitter Inherently imprecise due to dynamic changes in network topology and channel characteristics Routing decisions may not be accurate 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 4
Why QoS is hard in MANETs? 2/2 Lack of central coordination Error-prone shared medium Radio waves suffer from attenuation, multipath propagation and interference Hidden terminal problem Limited resource availability Bandwidth, battery life, storage, processing capabilities Insecure medium The broadcast nature of radio waves is highly insecure 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 5
Design Choices for QoS 1/2 QoS resource reservation is very important component for any QoS framework QoS reservation mechanisms classification Hard state vs. soft state Hard state Resources reserved at all intermediate nodes in path for duration of flow If path broken, resources must be explicitly released Requires control overhead May fail to release resources if nodes on path unreachable Soft state Resources reserved for small amount of time Reservations automatically renewed as long as flow continues Resources deallocated after timeout period if no new data No explicit tear-down needed Low overhead 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 6
Design Choices for QoS 2/2 Stateful vs. stateless Stateful Nodes keep either global or local state State includes topology information and flow information Global state not scalable Significant overhead incurred Local state is more accurate Simple distributed routing algorithm can be used Stateless No flow or topology information maintained at each node Scalable Difficult to provide QoS without knowing any state information MANETs specifics Dynamic, therefore hard QoS cannot be applied QoS guarantees can be given only within certain statistical bounds 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 7
QoS frameworks A framework for QoS A complete system that attempts to provide required/promised services to each user or application All components within the framework cooperate in providing the required services Key issue Serve on a per session basis Serve on a per class basis A class represents an aggregation of users based on certain criteria Serve on a per node basis Specific for MANETs 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 8
Key components of QoS in MANETS QoS architectures and models What type of services can be provided? Define the types of service differentiation QoS resource reservation signaling Coordinates routing, MAC, admission control and scheduling QoS routing Finds path with requested resources QoS MAC Provide support for QoS services 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 9
QoS architectures for the Internet Integrated Service (IntServ) Routers keep flow-specific state Bandwidth requirement Delay bound Flow cost Service models Best effort Guaranteed service: fixed delay bound Controlled load service: better then best effort RSVP protocol used to reserve resources in routers Admission control used to accept/decline reservations at hosts Priority queues implemented to provide service guarantees to flows with accepted reservations 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 10
IntServ for MANETs IntServ is not feasible in MANETs Scalability State information increases with number of flows Storage and processing overhead RSVP signaling packets use bandwidth needed to send data packets Burden on hosts Mobile hosts must perform admission control, classification of all incoming data packets, and priority scheduling 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 11
DiffServ architecture Differentiated Services (DiffServ) Featured by Traffic separated into small number of classes Routing decisions based on class of packet No per-flow state Scalable model Limited processing for routers Example services Premium service: low loss, low delay, low jitter, end-to-end bandwidth guarantee Assured Service: better than best effort service Olympic Service: three tiers of services Service Level Agreement (SLA) used to receive DiffServ Agreement between Customer and ISP ISP and ISP 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 12
DiffServ for MANETs No per-flow state ensures scalability DiffServ may be feasible for MANETs Premium service impossible to support Assured service possible SLA difficult to implement in MANETs 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 13
New QoS Model for MANETs: FQMM 1/2 Flexible QoS Model for MANET (FQMM) Hybrid approach Combines features from IntServ and DiffServ Per-flow QoS for high priority flows In FQMM, the nodes are classified into three categories Ingress node (source) Interior node (intermediate relay node) Egress node (destination) Ingress node responsible for traffic shaping Delaying packets belonging to flow to meet traffic profile Traffic profile Mean rate, burst size, etc. FQMM assumption The percentage of high-priority flows is less than of lower-priority ones FQMM QoS provision Provides per flow QoS for high-priority flows IntServ style Lower priority flows are aggregated into a set of service classes DiffServ style 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 14
New QoS Model for MANETs: FQMM 2/2 Issues with FQMM How many per-flow sessions possible? How do intermediate nodes determine packet information? How should scheduling be performed at intermediate nodes? New QoS models still needed for MANETs 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 15
QoS Signaling Used to reserve and release resources When flows created, removed or changed Informs application of success/failure of resource reservation Signaling protocol consists of three phases Connection establishment Connection maintenance Connection termination Two issues Reliable exchange of QoS signaling information In-band signaling: control information with data Out-of-band signaling: separate control packets Interpretation of QoS signaling information Protocols RSVP, Insignia 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 16
In-band vs. Out-of-band Signaling In-band signaling Low overhead Cannot implement complex functionality Out-of-band signaling Adds overhead Higher priority for signaling messages Reduces effective bandwidth for data transmission Easier to implement signaling protocol 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 17
RSVP QoS signaling for the Internet Out-of-band signaling system Request message sent via routing protocol to receiver Request includes traffic specifications (rate, burst size) Receiver sends back a reservation message to sender Intermediate routers check if they can support requested services If so, allocate resources If not, send error message to receiver Receiver initiates resource request Flow information periodically refreshed Problems for MANETS Too much overhead to apply RSVP to MANETs Not adaptive to dynamic networks 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 18
MRSVP 1/2 Extension of RSVP proposed for cellular networks to integrate with IP networks Predicts future locations and reserves resources Types of reservations Active A reservation over a path for a QoS flow is active if data packets currently flow along that path Passive A reservation over a path for a QoS flow is passive if the path on which resources have been reserved is to be used for data packets only in future Other flows can use resources from passive reservations MRSVP employs proxy agents To reserve resources along the path from the sender to the locations in the mobility specification of the mobile host The Local Proxy Agent Present at the current location of the mobile host Makes active reservations The Remote Proxy Agent Present at other locations in the mobility specification of the mobile host Makes passive reservations 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 19
MRSVP 2/2 Limitations MRSVP requires the future locations of mobile hosts in advance Obtaining of such information is difficult, mobile hosts future locations are unrestricted and therefore unpredictable Passive reservations fail in the case of MANETs Even if location is somehow known Finding a path and reserving resources on that path in advance may not be viable Because of random and unpredictable movement of the intermediate nodes It is unknown which nodes should act as proxy agents due to lack of infrastructure support 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 20
Insignia 1/3 QoS framework designed specifically for MANETs In-band signaling Adaptive services support, with different QoS level Base and enhanced QoS levels Base QoS level: minimum quantitative QoS Enhanced QoS level: extended base QoS level, when sufficient resources become available Key design issues in providing adaptive services How fast the application service level can be switched from base QoS to enhanced and vise versa, in response to changes in the network topology and channel conditions How and when it s possible to operate on the base QoS or enhanced QoS level for an adaptive applications It can scale down, drop, or scale up user sessions adaptively Based on network dynamics and user-supplied adaptation policies QoS reports periodically sent to source node Source node takes action to adapt flows to observed network conditions 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 21
Insignia 2/3 Key components Per-flow management Resource management adapted as topology changes Intelligent packet scheduling Flow reservation, restoration and adaptation Routing Any routing protocol can be used Route maintenance procedure will affect QoS In-band signaling Establish, adapt, tear down reservations Control information embedded in data packets Admission control Determine whether or not to accept reservation Refresh reservation periodically based on current state 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 22
Insignia 3/3 Key components Packet scheduling MAC Weighted round-robin for different flows Any MAC protocol can be used Automatic reconfirmation or de-allocation of reservation based on data packets received and timeouts Integrated in-band signaling, admission control and packet scheduling Useful for multimedia applications Support multiple operation modes (max and min bandwidth modes) Some loss acceptable MAC and routing protocols affect ability to support QoS 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 23
QoS Routing Protocols Goal Search for a path through the network that provides sufficient resources to meet QoS goals E.g., delay, delay jitter, bandwidth Concave or additive metrics for paths E.g., bandwidth is concave, whereby each link must satisfy minimum bandwidth constraints E.g., delay is additive, whereby route delay is sum of individual link delays NP-complete problem to find paths with two or more metrics E.g., finding delay-constrained least-cost path Difficulties with QoS routing Overhead Maintaining link state information Cannot guarantee QoS as in wired networks Route breaks Node failures Must update paths with new paths that have enough resources may not be possible 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 24
Ticket-based QoS Routing 1/2 Probe packets (PKT) issued with certain number of tickets Tickets determine maximum number of paths that can be probed for suitability Intermediate nodes allocate tickets among neighbors Neighbors are choose most likely satisfying QoS constraint Requires state information Example Source A transmits PKT with 3 tickets to neighbor B B transmits PKT with 2 tickets to C, PKT with 1 ticket to D C transmits PKT with 1 ticket to E, PKT with 1 ticket to F D transmits PKT with 1 ticket to G Etc. Multiple suitable paths found, select best one 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 25
Ticket-based QoS Routing 2/2 Reliability Multi-path routing Data sent independently on all paths Destination keeps first copy of data to arrive, discards rest Best path selected as primary path, others kept as back-up paths With resources reserved Without resources reserved best effort Provides nice trade-off between control overhead (based on number of tickets) and finding good feasible path Issues Performance depends on ticket-issuing and ticket-splitting procedures Global state required at each node Not scalable 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 26
QoS MAC Protocols Two basic approaches Guaranteed resource reservation Provide service differentiation Allow real-time/high priority packets to access channel before non-realtime/lower priority packets Still meet the goals of MAC protocols 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 27
Cluster TDMA Network organized into clusters Select cluster head via ID, etc. Cluster head assigns TDMA slots to nodes Inter-cluster interference avoided via TDMA or CDMA Frame times synchronized throughout network Create virtual connections via assigning slots Free slots used for best-effort traffic via slotted-aloha 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 28
MACA/PR MACA with Piggyback Reservation Data and ACK packets include reservation information in header Receivers keep reservation tables Bandwidth reservation Reservation information in Data packets inform neighbors of transmitter of next transmission Reservation information in ACK packets inform neighbors of receiver of next transmission Reservation tables also shared among nodes 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 29
Providing QoS in IEEE 802.11 Provide better than best effort service Service differentiation via Prioritization of different packets Fair scheduling Tunable parameters Contention window Backoff algorithm Interframe spacing (IFSs) 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 30
IEEE 802.11e Enhanced DCF (EDCF) All nodes use EDCF MAC protocol Parameters set according to traffic requirements Prioritizes traffic according to access category (AC) Adjust Minimum and maximum backoff window sizes Multiplication factor for adjusting backoff window Probability of accessing channel affected by these parameters Set parameters such that high priority data has higher probability of accessing channel earlier 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 31
QoS in IEEE 802.11 Other techniques for prioritization Persistent Factor DCF: backoff geometrically distributed with parameter P based on packet priority Fair-scheduling techniques Provide fairness in allocation of bandwidth to different traffic classes Often cannot be implemented in existing standard Distributed weighted fair queue Distributed fair scheduling Distributed deficit round robin 24-Nov-2009 TLT-2756 :: TUT :: fall 2009 :: Lecture 9 32