Facts About Soft RTS

(1)

Dr. Hugh Melvin, Dept. of IT, NUI,G

Soft - Firm RTS

Recall RTS Definitions

• A Soft RTS is one in which performance is

degraded but not destroyed by failure to meet

response time constraints (Laplante)

• Examples

– Typical Desktop PC

– Multimedia devices

• Eg. VoIP service i.e. public IP network – Unmanaged Înon deterministic

• Note: For MM data, requirement for logical correctness of output can be relaxed somewhat (See G.1010)

– Recall

• Unclear distinction between Soft and Firm RTS

Firm RTS

• A Firm RTS is one in which a few missed deadlines will

not lead to total failure, but missing more than a few may

lead to complete and catastrophic system failure failure

(Laplante)..

– Written with Hard RTS in mind..

• Distinction between Soft/Firm typically based on

consequences of missed deadlines

– Eg. Private IP Network governed by SLA • SLA may specify jitter/delay/loss/availability • Lack of adherence results in :

» Irate customers Î loss of business » Penalties imposed on provider

– Mobile Phone Technology / Cameras / Games Consoles etc… – E-Business eg. Reservation Systems

(2)

Soft – Firm RTS

• Embedded Systems

• Multimedia Systems (MMS)

• QoS is key issue for MMS

– QoS in Communication Networks • LAN based and WAN based MM

– Main focus for remainder of course

• LAN : QoS Support

– Integrated voice (media rich) /data systems

• WAN: QoS Support

– Many issues with public IP Networks

– Emergence of private IP networks with QoS Support

– QoS in MM Terminal

– Move towards RTOS functionality

– Other S/W and H/W hardware issues : Application Design, Driver performance, buffer design, skew

– ITU-T G.1010 Useful guideline

QoS

• Much abused term .. broad application

• Intrinsic versus Perceived QoS

– Intrinsic (Network) QoS

• Relates to technical performance of network – Packet loss

– Packet delay

– Packet jitter : variation in delay – (Above 3 influenced by bandwidth) • Other critical issues

– Availability (down time)

» POTS designed with 99.999% ‘line-available’ target – Connection setup time

QoS

• Perceived QoS: Same Intrinsic QoS may be

perceived very differently by different users

– Influenced by expectations and past experience

• Expectations rarely drop!

– Eg.

• Mobile phone technology

» Coverage, voice quality, availability expectations • Internet connectivity

» Bandwidth .. What does broadband mean? » What will it mean in 5 years?

» Contention rates often overlooked

» Narrowband (POTS-GSM) versus Wideband codecs » (eg. Skype can use both)

(3)

G.1010 Summary

T1213060-02 Fax Error tolerant Conversational voice and video

Voice/video messaging Streaming audio and video Error intolerant Command/control (e.g. Telnet, interactive games) Transactions (e.g. E-commerce, WWW browsing, Email access) Messaging, Downloads (e.g. FTP, still image)

Background (e.g. Usenet) Interactive (delay <<1 s) Responsive (delay ~2 s) Timely (delay ~10 s) Non-critical (delay >>10 s)

G.1010

• MM based Soft RTS primarily deal with

delay insensitive applications

– Often are Loss Tolerant

• Human comprehension can tolerate/compensate

for limited loss without significant QoS impact

• May also have strategies to deal with degree of

packet loss

– Basis for use of UDP versus TCP

• Note: Many streaming applications use TCP

» Delay not an issue as non interactive » Eg Internet Radio Service

(4)

Internet MM Protocols

RTP/RTCP (RFC 1889)

• Delivery of media streams across packet

network

• RTP

– Deals with some shortfalls of using UDP

– Sequence numbers

• o/o/o delivery • Detection of lost packets

– Media-specific timestamps

• 8000Hz typical for audio (POTS standard) • Used for reconstruction of media • Note: Silence Detection or VAD

– No packets sent during silences

– No assumption about sender/receiver clock synchronization

– Marker bit to indicate start of talkspurt

RTP Header

Internet MM Applications

• RTCP

– Companion Control Protocol to RTP

– RR (Recv Reports) and SR (Sender Reports)

– RR Provides feedback

• Packets lost / Jitter • Enables RTD to be calculated

– SR Facilitates Lip Synch

• Mapping between RTP timestamp and system clock time (NTP)

• Audio/Video streams must be tightly synchronised – Audio lead versus audio lag .. Which is best? • If NTP implemented

– Facilitates one way delay measurement – Skew elimination

(5)

Internet MM Applications

• Session Control eg. set-up/disconnect

– ITU-T H.323 or IETF SIP(Session Initiation Protocol)

• Alternative approaches • SIP gaining ground • Interoperability possible

– Messages carried over TCP

• TCP (control) plus UDP (media) channels

– Media Capability Issues

• Agree on codec for data transport

(6)

Packet Encapsulation

H.323 Based VoIP Session

VoIP

• Huge growth in recent years

• POTS Î VoIP

– Deterministic Synchronised Network Î Non

deterministic IP network

– Dumb ÎComplex Terminals

• Î Many challenges with both

• Streaming vs Interactive Audio

– Very different requirements

– VoIP: M2E delay critical

• G.114/G.1020 150 msec one-way limit • Good Echo cancellation facilitates higher delays

(7)

VoIP M2E Delay

• Sender

– Packetisation delay

– Encoding (+ look ahead) delays – OS/Applic/Driver s/w – NIC serialisation delay

• Network

– Propagation delay – Queuing delays

– Processing/serialisation delays in routers

• Receiver

– NIC delays – OS/Applic/Driver s/w – Jitter buffer delays – Decoding delays

VoIP QoS Strategies

• Sender-based

– RTCP feedback with adaptive codecs

• If loss/delay excessive, switch to lower b/w codec ? – FEC

• Network-based

– IntServ (RSVP), DiffServ, MPLS • Prioritising delay sensitive traffic flows

• Receiver-based buffering strategies

• Human ear very sensitive to short term variations Î Need to playout voice as close as possible to sampled voice • Buffer absorbs variation in network queuing delays • BUT.. adds to overall M2E delay

(8)

Jitter Buffer Strategies

• Buffer Playout Delay adds to M2E delay

• Above strategy

– Pkt 8 too late for playoutÎ normally dropped

–Î Increase size of buffer in response to increasing delays?

Jitter Buffer Strategies

• Fixed buffer size: limitations

• Too large Îextends overall delay (perhaps beyond G.114) • Too small ÎAdditional late packet losses due to late arrival

• Adaptive buffer size

• Adapt to network conditions • Per talkspurt or per packet

• Mostly operate by elongating/shortening inter talkspurt silence periods

–Îless noticeable

(9)

Network QoS

• LAN Environment

– Huge growth in integrated voice/data networks

• WAN Environment

– Public IP i.e. Internet

– Private IP networks

• ATM based solution

– Never succeeded in LAN environment

– WAN Deployment

ATM System Architecture

• Adaptation Layer (AAL): Inserts/extracts information into

48 byte payload

• ATM Layer: Adds/removes 5 byte header to payload

• Physical Layer: Converts to appropriate electrical or optical

format

End Station Switch End Station

Voice Voice Data Data Video Video

A

L

P

H

Y

P

H

Y

A

T

M

P

H

Y

A

T

M

P

H

Y

A

T

M

A

L

Voice Data Video Cells

QoS in LAN Environment

• Integrated voice / data services across Enterprise LANs

– Original Ethernet • Non deterministic

• Poor performance under high loads – Switched LANs

• Full duplex • High Speed Backplane • RTOS Engine

– Deterministic & QoS Support

• Cost effective

• Facilitates new generation of applications

– Collaborative workspaces : voice + video + ..etc..

• Better speech quality possible via wideband voice codecs • Gateway enables connection to POTS

(10)

CSMA-CD

(11)

QoS in LAN Environment

• Switched LANs typically QoS enabled

• Classify traffic according to :

• TL port number or Protocol (TCP vs UDP) • NL IP address, VLAN

• DLL MAC address • Physical port number

• Specify treatment for each classification

• Simple priorities • Bandwidth min & max levels

• Eg. UDP traffic (i.e. voice) from designated ports guaranteed 2% min traffic

– 2% of 100Mbps= 2Mbps

» Recall PCM G.711 voice = 64kbps

QoS in LAN Environment

• VLAN

– Logical rather than physical grouping of machines • VLANs used to reflect organisational structure • Easy to reconfigure

• Facilitates greater security • Limits broadcast traffic (ARP)

» Contained within each VLAN • Less scope for active attacks

• VLAN QoS

• May be configured according to VLAN grouping VLANs can span multiple switches and physical locations • Configuration

– By switch port / MAC address / IP address

VLAN

• IEEE 802.1Q (1998)

– Simplifies multi switch environment

• Each switch needs to know VLAN membership of nodes on other switches

– VLAN tag

• Extra 4 byte header field • Includes VLAN Identifier

– Useful to identify VLAN members across multiple switches – Becoming more prevalent

• Note: Fast/Gigabit Ethernet links rarely congested

– No need (yet) for QoS policies? – “What Andy giveth.. Bill taketh away”!

(12)

Other LAN Based Protocols

• Token Bus Protocol IEEE 802.4

– Token circulates around logical ring

– Node can only transmit when in possession of token

– Different priority levels

– Developed for production line environment

– Significant improvement on standard Ethernet 802.3

– Basis for Moneypoint CS275 Bus

• Upgrades: Switched LAN provides similar capability

• Largely superceded

QoS in WAN Environment

• Internet default ‘best-effort’ service

– TCP provides a reliable bit stream • Connection oriented

• Reliable service • Flow control – but..

• No guarantees on available bandwidth / loss / delay / jitter • TCP adds significant overhead and delay (for retransmissions) • Non deterministic

– Core Issue for realtime applications »Î use of UDP to limit delay

• Various initiatives to provide IP-based WAN QoS

– Distinction between Internet and private (managed) IP networks

(13)

QoS Approaches

• Increase bandwidth

– LAN solution: • Fast / Gigabit Ethernet

– No real need yet for QoS features?

– Last mile: Home Environment • ADSL / 3G

– Overprovision networks

• Costly, limited and temporary solution ?

• Reservation policy

– Similar to circuit switched POTS approach – Implementation / scalability issues – Requires admission control policy

• Traffic categorisation & prioritisation

– Simpler to implement

(14)

IntServ

• IETF IntServ (Integrated Services)

– RSVP (Resource Reservation Protocol rfc 2205)

– Resources reserved along path prior to media

delivery

• Routers need to maintain this information for duration of session

• Similar to virtual circuit

– Guaranteed and Controlled load service

• Guaranteed Î tight adherence to specification • Controlled load Î service equiv to lightly loaded network

DiffServ

• IETF DiffServ (RFC 2475)

– Prioritise traffic

– Traffic classification and policing carried out at edge

of network

– Recall IPv4 TOS and IPv6 Traffic Class fields

• Renamed DS Codepoint field (DSCP)

• DSCP interpreted by routers in terms of PHB (per hop behaviour)

• Simpler to implement within network

– PHB:

• Expedited Forwarding : Service equiv to a leased line – Realtime traffic has reserved bandwidth eg. VoIP • Assured Forwarding: 4 priority classes each with 3 different

drop precedence values

DiffServ EF PHB

(15)

DiffServ AF PHB

• Classifier

– Packets sorted into 4 priority classes – Sorting may be based on

• IP address

• Transport Layer Protocol etc.

• Marker

– Packet DSCP marked to indicate priority level

• Shaper/Dropper

– Shapes traffic flows into acceptable profile – May be governed by a Service Level Agreement

• Packets proceed into DiffServ aware network

DiffServ AF PHB

MPLS

• MPLS : Multi-Protocol Label Switching (RFC 3031)

• New switching architecture rather than purely QoS

focus

• Works with Frame Relay / ATM or IP

– Sometimes known as level 2.5

• Labels packets according to FEC (Forwarding

Equivalent Class)

– FEC determined at point of ingress to network

– New label attached at each router which will be correctly

interpreted at next downstream router

(16)

MPLS

Packet Loss Strategies

• In absence of Network QoS

– Use of UDP limits delays but leads to packet loss –Î Need to compensate

– Use of FEC & PLC

• Forward Error Correction (FEC)

– Form of Information Redundancy • Media Specific FEC

– Lower bandwidth version sent » Lower Quality – Eg. Dual codec capability

• Media Indep FEC: Exact replica sent

– Packet n also sent with Pkt n+1, n+2 etc » If Pkt n lost, must wait for Pkt n+1.. Or n+2

• Both strategies deliver correctness at cost of increased delay and bandwidth utilisation