Integrated Service in the Internet Architecture RFC 1633.

Integrated Service in the Internet Architecture

RFC 1633

Introduction

The Internet only offers simple QoS (quality of service)—best effortReal-time applications do not work well across the Internet because of: Variable queueing delays Congestion losses

The Internet infrastructure must be modified to support real-time QoS

Introduction

Real-time QoS is the issue for a next generation of traffic management in the InternetThe term integrated services(IS) for an Internet service model includes: Best-effort service Real-time service Controlled link sharing

Elements of the Architecture

The fundamental service model of the Internet—best effort has been unchanged for 20 yearsChange the service model of the Internet is a major undertakingNew components will supplement but not replace the basic IP serviceOnly to extend the original architecture

Integrated Service Model

Two sorts of service targeted towards real-time traffic: Guaranteed service Predictive service

It integrate with controlled link-sharingThe resources (e.g., bandwidth) must be explicitly managed

The arguments against resource guarantees

Bandwidth will be infinite In the future, the bandwidth will be so

abundant, ubiquitous, and cheap?

Simple priority is sufficient Simply giving higher priority to real-

time traffic is enough?

Applications can adapt

Integrated Service Model

There is an inescapable requirement for routers to be able to reserve resourcesProvide special QoS for specific user packet streams, or flowsUse the existing internet-layer protocol (e.g., IP or CLNP) for real-time data

Reference Implementation Framework

Propose a reference implementation framework to realize the IS modelThe framework includes 4 components: Packet scheduler Admission control Classifier Reservation setup protocol

Traffic control

For integrated services, a router must implement an appropriate QoS for each flowThe router function that creates different qualities of service is called “traffic control” Implemented by: the packet scheduler, the classifier and admission control

Traffic control

Packet Scheduler An experimental scheduler—CSZ

scheduler

Classifier Packets are mapped into some classes Packets in same class get the same

treatment form packet scheduler

Admission Control The decision algorithm used by router

The 4th component—reservation setup protocol

Create and maintain flow-specific state in the endpoint hosts and in routers along the path of a flowRSVP (ReSerVation Protocol) is used to reserve the resource

Implementation Reference Model for Routers

Routing Agent

Reservation Setup Agent

Management Agent

Admission Control

[ Routing ][ Database ]

[ Traffic Control Database ]

==========================================

Classifier Packet Scheduler

Input Driver

Internet Forwarder Output Driver

Implementation Reference Model for Routers

The forwarding path is divided into 3 sections : Input driver,internet forwarder,output

driver

Internet forwarder interprets the internetworking protocol header (e.g., IP header for TCP/IP)The output driver implements the packet scheduler

Implementation Reference Model for Routers

In routers, integrated service will require changes to both the forwarding path and the background functionsThe forwarding path may depend upon hardware acceleration for performance—difficult and costly to change

Quality of Service Requirements

Per-packet delay is the central quantity about which the network makes QoS commitmentsReal-time applications Need the data in each packet by a

certain time, or the data will be worthless

Elastic applications Always wait for data to arrive

Real-time applications

Playback applications The source takes some signal,

packetizes it, and then transmits over the network

Receiver has to buffer the incoming data and then replay the signal at some fixed offset delay form the original departure time

The performance is measured by Latency and fidelity

Real-time applications

Delay can affect the performance of playback applications in two ways: The value of the offset delay The delays of individual packets can

decrease the fidelity of the playback by exceeding the offset delay

Real-time applications

Intolerant applications Must use a fixed offset delay Set the upper bound on max delay Be called as guaranteed service

Tolerant applications Can tolerate some late packets Vary offset delays according to the

experience in the recent past Be called as predictive service

Elastic applications

Always wait for data to arriveExample applications: Interactive burst — Telnet Interactive bulk burst — FTP Asynchronous bulk transfer — E-mail

An appropriate service model for these applications is to provide as-soon-as-possible service (i.e., best-effort service)

Resource-sharing requirements

Multi-entity link-sharing When the link is underloaded, any one

of the entities could utilize all idle bandwidth

Multi-protocol link-sharing Prevent one protocol family from

overloading the link

Multi-service sharing Limit the amount real-time traffic to

avoid preempting elastic traffic

Other remarks

Packet dropping Some of the packet within a flow

could be marked as preemptable Router use this mark to drop packets

Usage feedback Prevent abuse of network resources

Reservation model Describe how an application

negotiates for a QoS level

Traffic Control Mechanisms

Basic functions: Packet scheduling Packet dropping Packet classification Admission control

An example: The CSZ scheme

Packet scheduling

Reorder the output queueOne approach is a priority scheme Packets are ordered by priority Highest priority packets leave first

An alternative scheme is round-robin Gives different classes of packets

access to a share of the link

Packet dropping

A router must drop packets when its buffers are all fullDropping the arriving packet is simple but may cause undesired behaviorIn real-time service, dropping one packet will reduce the delay of all the packet behind itDropping and scheduling must be coordinated

Packet classificationThe classifier implementation issues are complexity and processing overheadOne approach is to provide a flow-id field in the Internet-layer packet header Reduce the overhead of classification

Engineering is required to choose the best design of this concept

Admission control

Admission control—the design about resource availabilityThe router has to understand the demands that are currently being made on its assetsA recent proposal is to program the router to measure the actual usage by existing packet flows, then use this information for the admitting of new flow

The CSZ scheme

At the top level, CSZ node use WFQ to separate guaranteed flows for each otherPredictive and best-effort service are separated by priorityInside each predictive sub-class, FIFO queueing is used to mix the traffic

The CSZ scheme

Within the best-effort class, WFQ is used to provide link sharingWithin each link share of the best-effort class, priority is used to permit more time-sensitive elastic trafficThe CSZ node uses both WFQ and priority in an alternating manner to build the mechanism

Reservation Setup Protocol

Requirements for the design of a reservation setup protocol: designed for a multicast environment accommodate heterogeneous service needs can add/delete one sender/receiver to an

existing set robust and scale well to large multicast

groups advanced reservation of resources, and for

the preemption

RSVPFlowspecs and Filter Specs RSVP reservation request specifies

the amount of resources to be reserved

The resource quantity is specified by a flowspec

The packet subset to receive those resources is specified by a filter spec

The service model presented to an app. must specify how to encode flowspecs and filter specs

RSVP—reservation styles

Offers several different reservation styles Wildcard

All packet destined for the session may use a common pool of reserved resource

Fixed-filter Can not be changed during its life time

without re-invoking admission control Dynamic-filter

Receiver can modify its choice of resource without additional admission control

RSVP—reservation styles

Wildcard uses a filter spec that is not source-specificThe other two use filter specs that select particular sourcesThe wildcard reservation is useful in support of an audio conference

RSVP—initiation

Sender knows the qualities of the traffic stream it can sendReceiver knows what it wants to (or can) receiveSender initiation scales poorly for large, dynamic multicast delivery trees and for heterogeneous receiversThus, RSVP uses Receiver-Initiation

RSVP—initiation

Receiver Initiation Natural choice for multicast sessions But may appear weaker for unicast

sessions

Except real-time app. will have its higher-level signalling and protocolThen this protocol can be used to signal the receiver to initiate a reservation

RSVP—states

Hard state approach Connection-oriented

Soft state approach Connectionless

RSVP takes the Soft State approach Regards the reservation as cached

information that is installed and periodically refreshed by the end hosts

RSVP—routing issues

Find a route that support resource reservationFind a route that has sufficient unreserved capacity for new flowAdapt to a route failureAdapt to a route change (without failure)The last issue is provide by mobile hosts

Conclusion

The Integrated services framework has four main components : Packet scheduler Admission control Classifier Reservation setup protocol

RSVP is used to reserve the resource for the session belongs to high class level