Advanced networking - scheduling and QoS part 1

Scheduling and Quality of Services (QoS)

Advanced Telecommunication Network(ET5187)

byAris Cahyadi Risdianto

23210016

Scheduling and QoS

Input ===> ===> Output (Controller)

Scheduling and QoS

Packet Classification

• Same Class / No Class differentiation > FIFO/LIFO > Most Common Used

• Different Class > Lost Sensitive and Delay Sensitive > Different rules for different Class

Queuing System

I = class of service of K and J flowsMi(t) = actual service allocated for class i at t timeNi(t) = Buffer size

Loss Sensitive Scheduling

Two different Class (High and Low), for the each same class use FIFO with K buffer• Head of line (HoL)• Partial Buffer Sharing (PBS)• Push Out Buffer (POB)• Random Early Detection (RED)

Buffer Size

Familiar and Famous for ATM network

D* = Delay constraint end to end (Based on ITU) -> 10 msE(H) = average number hops ( <10)M = maximum service rate for STM-1 155 Mbps = 366800 cellsK <= M x D*/10 ms

K small, assured delay but loss cell

Head of Line

• Known as priority queue for queueu > 2• Always served High priority cells in the buffer• pre-emptive and non pre-empetive

Same as :

H3 | L5 | L4 | H2 | L3 | H1 | L2 | L1 => HoL

L5 | L4 | L3 | L2 | L1 | H3 | H2 | H1 => FIFO

Partial Buffer Sharing

• The rule specified by threshold T in the queue• Nq(t) is number cells/packet at the time T• Nq(t) < T, high and low enter the queue• Nq(t) > T, only high enter the queue• Nq(t) = K, buffer is full, arriving cells discarded

Push Out Buffer

• Avoid complex determination of best position• Only operates when the buffer is full

LIFO FOB R FOB FIFO FOB

H3 ==> | L5 | L4 | H2 | L3 | H1 | L2 | L1

Random Early Detection (RED)

• Like PBS but have 2 threshold Tmin and Tmax• q < Tmin => no packet drop• q > Tmax => all packet are dropped• Tmin < q < Tmax => packet are drop with P = (q-

Tmin) * Pmx / Tmax-Tmin• used for TCP flows congestion avoidance• variant of RED is WRED (Weighted RED)

Delay Sensitive Scheduling

• Assumption there is no problem of losing packet and buffer k is long

• Consist of processor sharing example WRR has class k with weight Wk

• WRR (Processing sharing) inflexible because Wk independent

• Processing rule rather than processing sharing

Upper Bound Method

• Used for solving CAC (Call Admission Control) problem

• Some assumption : o Each arrival process satisfies with certain

business constraino Service time for cell/packet is deterministic and

proportionalo Scheduling rule is used to generate QoS for

class k with minimal Mk ("fair" rule to prevent blocking another class getting served)

Upper Bound Method (Cont.)

• Queue count is maximum difference between inflow and outflow (λk and μk)

• If queue > 0, class served by minimal rate (μk)• Number of queue bounded by burstinest σk

provided if λk ≤ μk• Buffer size bounded by sum of burstinest all flows,

so loss can be guaranteed• Maximum delay bounded by burstinest divide by

inflows, so delay can be guaranteed

Upper Bound Method (Cont.)

• Remarks on upper bound method :• Zero packet loss only guaranteed for admitted

packet (satisfied with burstinest constrain), if not packet will be lost

• Delay guaranteed are deterministic because all stochastic assumed to be bounded or deterministic

• Upper Bound Method more optimal than N*D/D/1 queuing for scenario where N not identical and independent CBR resources

QoS (Quality of Services)

Evolution and Importance

• Internet and Value Added Services is the main driver

• Internet used for e-commerce, self backing and communication

• Overall result : people are tolerant about QoS, in certain point some people are frustrated of losing data

• Mature Internet need to offer ubiquitous inexpensive, and high quality services

QoS as Technological Lever

• Two Main Approach :

o Over Installing resources (less than 30% load)o Controlling Traffic in the network to ensure each

flows achieve certain level of QoS• QoS implementation is faster and cost effective

than expanding new network (Fiber, equipment, etc)

QoS as Commercial Lever

• Old view: over dimension network without complex network functionality

• Notation QoS : demand always beyond supply cause congestion

• QoS offer dividing resources, not guarantee quality (lower priority users get less, high priority users get more)

• Controlling QoS = Controlling Resources• Sub-optimal controlling resource = loss revenue

Definition and Property of QoS

• In general, QoS express set of service example performance, availability, reliability and security

• Network QoS on layer 3 (inspired by ATM network)

• Application Layer QoS associate with GoS• QoS provision cause dichotomy “soft” and “hard guarantee”

• Perceived QoS (Voice, Streaming, e-games): delay, jitter, echo, packet loss

Challenges : QoS aware Networking

• Main Problem : Stochastic arrival process and deterministic set of traffic or determination

• Analysis and computing random variable more complex

• Computation level explode state spaces and prevent accurate computations

• Inherent problem “connection set-up time” need QoS per flow such as QoS routing, signalling and CAC

• Two different future QoS by IETF : IntServ and DiffServ

Evolution Network Architecture : IP and ATM

• Two different approach : IP by IETF and ATM by

ATM forum• Layered routing : IP for L3 and ATM for L2• Integration :

• Partially Integrated : Dual-Mode• Fully Integrated : I-PNNI• Ipsilon : ATM for forwarding, IP for control• IETF : MPLS (IP Fast Switching)

• ATM goal, reality of IP : Basic Architecture for Broadband Multimedia

QoS Emerging in the Internet

• QoS aware networking including QoS routing, signaling and traffic management

• Standardized by IETF but not implemented• QoS aware Internet:

• RSVP : signaling• IntServ : end-to-end signaling per flow basis• DiffServ : no end-to-end signaling per flow

RSVP

• RSVP (Resource Reservation Protocol)• IETF signalling protocol based on multicast• Used two types of messages : path and reservation

messages• Most of Telecom or connection oriented based on

unicast• Continued with SIP protocol as application layer

signalling protocol

RSVP Messages

• Path messages : previous hop IP address, sender template and IP address, traffic characteristic, end to end QoS requirement

• Teardown Messages : Path Tear and Resv Tearo Path Tear : Iniated by the sender to install

reservation stateo Reservation Tear : Travels from receiver to the

sender to remove reservation state

RSVP Operations

• Sender send path message to receiver of the mcast group, each router install the reservation state and record the hop

• Receiver send "Resv Message" to nearest router and ask amount of resources

• Nearest Router reserve along the path to the sender

• If other receiver joint the mcast group, nearest router ask more resource along the path

Characteristic of RSVP

• Used for unicast and multicast application• Receiver Oriented : flows initiated and resources

reservation• Consist of policy control and admission control• MPLS LDP is alternate to RSVP based on explicit

routing• QoS state is soft state : messages are flows

periodic to adopt the routing changes• reservation for unidirectional data flows

Integrated Services (IntServ)

• Additional component : packet classifier, scheduler and admission control

• Required resources reservation for each session/flows using RSVP

• If RSVP failed, the session will be best effort (BE)• Two class:

o Guaranteed Services : provide services with guaranteed both delay and bandwidth

o Controlled-Load Services : provide data flow all same in the unloaded network, and using CAC if network is loaded

Differentiated Services (DiffServ)

• Threat each class differently on per-hop behaviour (PHB)

• Class differentiation rather than flow differentiation (more scalable)

• Provide QoS more natural than IntServ which inline with Internet

• Bandwidth Broker use to managed inter-domain resources for providing end-to-end QoS

Differentiated Class

• IP DSCP format:

• Two different PHB Class, except BE (Best Effort) : Expedited Forwarding (EF) = virtual leased line or

point-to-point connection Assured Forwarding (AF) = better best efforf

Next on "Scheduling"

• Generalized Processing Sharing (GPS)• Generalized cu-rules (Dynamic Scheduling

Rules)

Next on "Quality of Services"

• Shortcut Routing : L2 over L3 (MPLS)• Multiprotocol Label Switching (MPLS)

including GMPLS (Generalized MPLS)