Rtp Slides

RTP

1

Real-Time Transport Protocol (RTP)

August 12, 2001

RTP

2

RTP • protocol goals • mixers and translators • control: awareness, QOS feedback • media adaptation

August 12, 2001

RTP

3

RTP – the big picture

application

media encapsulation

RTP RTCP

data

control UDP

ST-II

IPv4/6

Ethernet

AAL5 ATM

August 12, 2001

RTP

4

RTP = Real-time transport protocol • only part of puzzle: reservations, OS, . . . • product of Internet Engineering Task Force, AVT WG • RFC 1889, 1890 (to be revised) • initiated by ITU H.323 (conferencing, Internet telephony), RTSP, SIP, . . . • support for functions, but does not restrict implementation • compression for low-bandwidth networks: CRTP (RFC 2508)

August 12, 2001

RTP

5

RTP goals lightweight: specification and implementation flexible: provide mechanism, don’t dictate algorithms protocol-neutral: UDP/IP, ST-II, IPX, ATM-AALx, . . . scalable: unicast, multicast from 2 to O(107 ) separate control/data: some functions may be taken over by conference control protocol secure: support for encryption, possibly authentication

August 12, 2001

RTP

6

Data transport – RTP Real-Time Transport Protocol (RTP) = data + control data: timing, loss detection, content labeling, talkspurts, encryption control: (RTCP) ➠ periodic with T ∼ population • QOS feedback • membership estimation • loop detection

August 12, 2001

RTP

7

RTP functions • segmentation/reassembly done by UDP (or similar) • resequencing (if needed) • loss detection for quality estimation, recovery • intra-media synchronization: remove delay jitter through playout buffer • intra-media synchronization: drifting sampling clocks • inter-media synchronization (lip sync between audio and video) • quality-of-service feedback and rate adaptation • source identification

August 12, 2001

RTP

8

RTP mixers, translators, . . . mixer: • several media stream ➠ one new stream (new encoding) • mixer: reduced bandwidth networks (dial-up) • appears as new source, with own identifier translator: • single media stream • may convert encoding • protocol translation (native ATM ↔ IP), firewall • all packets: source address = translator address

August 12, 2001

RTP

9

RTP mixers, translators, . . .

end system SSRC=17 DVI4 192.35.149.52

end system SSRC=39

L16

translator 192.26.8.84

GSM

mixer

GSM

SSRC=5

SSRC=5 CSRC= 17 39

192.20.225.101

128.119.40.186

August 12, 2001

RTP

10

RTP packet header bit 0

8

16

V(2)P X CSRC count M payload type

24

32

sequence number

contributing source identifiers (CSRC) header extension

opt. opt.

synchronization source identifier (SSRC)

payload (audio,video,...) 0x00

opt.

UDP packet

timestamp

bytes August 12, 2001

RTP

11

RTP packet header Payload type: audio/video encoding method; may change during session SSRC: sychronization source ➠ sources pick at random ➠ may change after collision! sequence number: +1 each packet ➠ gaps ≡ loss P: padding (for encryption) ➠ last byte has padding count M: marker bit; frame, start of talkspurt ➠ delay adjustment CC: content source count (for mixers) CSRC: identifiers of those contributing to (mixed into) packet

August 12, 2001

RTP

12

RTP timestamp • +1 per sample (e.g., 160 for 20 ms packets @ 8000 Hz) • random starting value • different fixed rate for each audio PT • 90 kHz for video • several video frames may have same timestamp • ➠ gaps ≡ silence • time per packet may vary • split video frame (carefully. . . ) across packets • typical: 20 to 100 ms of audio

August 12, 2001

RTP

13

RTP in a network • typical: UDP, no fixed port; RTCP port = RTP port (even) + 1 • typical UDP size limited to few hundred bytes (OS, network, fragmentation) • native ATM: directly into AAL5 frame • encapsulation (length field) for others • typically: one media (audio, video, . . . ) per port pair • exception: bundled MPEG

August 12, 2001

RTP

14

RTP control protocol – types stackable packets, similar to data packets sender report (SR): bytes send ➠ estimate rate; timestamp ➠ synchronization reception reports (RR): number of packets sent and expected ➠ loss, interarrival jitter, round-trip delay source description (SDES): name, email, location, . . . CNAME (canonical name = user@host) identifies user across media explicit leave (BYE): in addition to time-out extensions (APP): application-specific (none yet)

August 12, 2001

RTP

15

RTCP packet structure if encrypted: random 32-bit integer packet

packet

chunk SDES

item

CNAME PHONE

item

SSRC

site 2

item

SSRC

site 1

SSRC

sender report

SSRC

SR

SSRC

receiver reports

chunk CNAME

item LOC

BYE

SSRC SSRC

packet

reason

compound packet UDP packet

August 12, 2001

RTP

16

RTCP announcement interval computation Goals: • estimate current # & identities of participants – dynamic • source description (“SDES”) ➠ who’s talking? • quality-of-service feedback ➠ adjust sender rate • to O(1000) participants, few % of data ➠ randomized response with rate ↓ as members ↑ • group size limited by tolerable age of status • gives active senders more bandwidth • soft state: delete if silent

August 12, 2001

RTP

17

RTCP bandwidth scaling • every participant: periodically multicast RTCP packet to same group as data • ➠ everybody knows (eventually) who’s out there • session bandwidth: – single audio stream P – of concurrently active video streams

August 12, 2001

RTP

18

RTCP bandwidth scaling • sender period T : T =

# of senders · avg. RTCP packet size 0.25 · 0.05 · session bw

• receivers: # of receivers · avg. RTCP packet size T = 0.75 · 0.05 · session bw • next packet = last packet + max(5 s, T ) · random(0.5. . . 1.5) • randomization prevents “bunching” • to reduce RTCP bandwidth, alternate between SDES components

August 12, 2001

RTP

19

RTCP sender reports (SR) SSRC of sender: identifies source of data NTP timestamp: when report was sent RTP timestamp: corresponding “RTP time” ➠ lip sync sender’s packet count: total number sent sender’s octet count: total number sent followed by zero or more receiver report

August 12, 2001

RTP

20

RTCP receiver reports (RR) SSRC of source: identifies who’s being reported on fraction lost: binary fraction cumulative number of packets lost: long-term loss highest sequence number received: compare losses, disconnect interarrival jitter: smoothed interpacket distortion LSR: time last SR heard DLSR: delay since last SR

August 12, 2001

RTP

21

Intermedia synchronization = sync different streams (audio, video, slides, . . . ) • timestamps are offset with random intervals • may not tick at nominal rate • SRs correlate “real” time (wallclock time) with RTP ts 560 = 8:45:17.23 audio 0 RTP video

160

320

480 640 RTCP SR

800

960

RTP timestamp

1800 = 8:45:17.18 0

RTCP SR

1120

9000

August 12, 2001

RTP

22

Round-trip delay estimation compute round-trip delay between data sender and receiver [10 Nov 2001 11:33:25.125] n

SR(n)

[10 Nov 2001 11:33:36.5]

A=0xb710:8000 (46864.500 s)

ntp_sec =0xb44db705 ntp_frac=0x20000000 (3024992016.125 s)

dlsr=0x0005.4000 ( 5.250 s) lsr =0xb705:2000 (46853.125 s) RR(n)

r DLSR (5.25 s) A 0xb710:8000 (46864.500 s) DLSR −0x0005:4000 ( 5.250 s) LSR −0xb705:2000 (46853.125 s) delay 0x

6:2000 (

6.125 s)

August 12, 2001

RTP

23

RTP: Large groups How do manage large groups? • “movie at ten” • channel surfing ➠ reconsideration: pause and recompute interval • conditional reconsideration: only if group size estimate increases • unconditional reconsideration: always • reverse reconsideration to avoid time-outs

August 12, 2001

RTP

24

BYE floods • avoid BYE floods: don’t send BYE if no RTCP • reconsideration More general: • general bandwidth sharing problem • “squeaky wheel” network management

August 12, 2001

RTP

25

Reconsideration: learning curve Learning Curve 100000

10000

Number

1000

100

L(t) Current L(t) Conditional Reconsideration L(t) Unconditional Reconsideration Ideal

10

1 1

10

100

1000

10000

100000

Time (s)

August 12, 2001

RTP

26

Reconsideration: influence of delay Cumulative Packets Sent 3000 Uniform Fixed Exponential 2500

Number

2000

1500

1000

500

0 0

100

200

300

400

500 Time (s)

600

700

800

900

1000

August 12, 2001

RTP

27

RTP: Aggregation • interconnected IPTel gateways ➠ several RTP streams to same destination • high overhead: G.729, 30 ms packetization ➠ 30 bytes audio, 40 bytes IP + UDP + RTP headers • with ATM: efficiency = 28% • solution: bundle several calls into single RTP session

August 12, 2001

RTP

28

RTP: Aggregation M

sequence number

PT timestamp SSRC

M

PT

0

call ID

M

PT

1

call ID

M

PT

0

call ID

length of block

first audio block

second audio block

third audio block

• for 24 channels ➠ efficiency ↑ 89% • signal call-ID using SIP August 12, 2001

RTP

29

Collision detection and resolution Collision: • two sources may pick the same SSRC (“birthday problem”) • probability: about 10−4 if 1000 session members join more or less simultaneously • but: don’t pick one you know about already ➠ probability much lower unless everyone joins at the same time • send BYE for old, pick a new identifier

August 12, 2001

RTP

30

Loop detection • forward packet to same multicast group (directly or through translators) • looks similar to collision, but changing SSRC doesn’t help • look at RTCP packets

August 12, 2001

RTP

31

RTP for the masses • for 14.4 kb/s stream: 90 B/s ≈ 1 new site/s • takes ≈ 3 hours to get to know 10,000 people ➠ – who cares? (Nielsen!) – useless for QOS feedback – control rate too high • ➠ statistical sample (sender determines rate): send value [0, 1]; pick random value; if <, lucky winner ➠ needs to be adaptive • ➠ report just to sender, instead of multicast

August 12, 2001

RTP

32

Adaptive applications

August 12, 2001

RTP

33

Adaptive applications Multimedia applications can adjust their data rates: Audio: encoding parameters (MPEG L3), encoding, sampling rate, mono/stereo encoding sampling rate bit rate LPC 8,000 5,600 GSM 8,000 13,200 DVI4 8,000 32,000 µ-law 8,000 64,000 DVI4 16,000 64,000 a range of DVI4 and MPEG L3 L16 stereo 44,100 1,411,200

August 12, 2001

RTP

34

Adaptive applications Video: frame rate, quantization, image resolution, encoding 30000 Noise Face 1 Face 2 White Picture Black Picture

Size [Bytes]

25000 20000 15000 10000 5000 0 0

50

100

150 200 Q-Factor

250

300

August 12, 2001

RTP

35

Application control • networks with QoS guarantees: – QoS at call set-up, guaranteed – long call durations ➠ network load may change – “wrong” guess ➠ rejected calls or low quality • networks w/o QoS or shared reserved link: – adapt application to available bandwidth – share bandwidth fairly with TCP? – lowest common demoninator ➠ mixers, translators

August 12, 2001

RTP

36

TCP-friendly applications • avoid race due to FEC, aggressive retransmission • push aside TCP applications (sometimes ok. . . ) • avoid congestion collapse • avoid being but in “penalty box” • time scale?

August 12, 2001

RTP

37

TCP-friendly adaptation • rate computation (e.g.,): – use additive-increase, multiplicative-decrease √ , where R is the round-trip time and – use loss/RTT equation: throughput = R1.22 p p ≈ loss fraction • mechanisms: – TCP ACKs, without retransmission −→ overhead, no multicast – RTCP RR −→ delay, metric?

August 12, 2001

RTP

38

RTP: Status and Issues Compression: differential compression for low-speed point-to-point links ➠ compress IP, UDP, RTP into 1–2 bytes Aggregation: trunking of packet streams or Internet telephony gateways Large groups: RTCP feedback for O(10,000); sampling RTP (RFC 1889, RFC 1890) −→ draft standard

August 12, 2001

RTP

39

RTP Header Compression • large overhead for IP + UDP + RTP headers: 40 bytes • CRTP = lossless differential compression that reduces overhead to two bytes on (low-speed) point-to-point links • derived from VJ TCP/IP header compression • context: IP address, port, RTP SSRC • IP: only packet ID changes • UDP: only checksum • RTP: second-order difference of timestamp and sequence number is zero • resynchronization by NAK −→ not good for high BER, delay

August 12, 2001

RTP

40

RTP Header Compression • link layer indicates FULL_HEADER, COMPRESSED_UDP, COMPRESSED_RTP, CONTEXT_STATE (no IP header) • differences are encoded as variable-length fields: -16384 -129 -128 -1 0 127 128 16383 16384 4194303

C0 00 00 C0 3F 7F 80 00 80 7F 00 7F 80 80 BF FF C0 40 00 FF FF FF August 12, 2001

RTP

41

CRTP Packet Header 0

1

2

3

4

5

6

7

msb of session context ID (if 16−bit CID)

lsb of session context ID M (marker)

S

T

I

link sequence

UDP checksum (if non−zero in context)

random fields (if non−zero in context) [used when IP−in−IP header]

if MSTI = 1111

M’

S’

T’

I’

CC (CSRC count)

delta IPv4 ID (if I or I’ = 1)

delta RTP sequence (if S or S’ = 1)

delta RTP timestamp (if T or T’ = 1)

CSRC list (if MSTI = 1111 and CC != 0)

RTP header extension (if X bit set in context)

RTP data

August 12, 2001

RTP

42

Some RTP Implementations tool NeVoT vic vat rat Rendezvous NetMeeting IP/TV RM G2

who GMD Fokus LBNL LBNL UCL INRIA Microsoft Cisco Real

media audio video audio audio A/V A/V A/V A/V

RSVP yes no no no no no no no

adaptive not yet no no no yes no no yes

http://www.cs.columbia.edu/˜hgs/rtp/

August 12, 2001