26
Speaker: 邱邱邱 Date:2010/10/4 Using sigcomp to compress SIP/SDP Messages Haipeng Jin and AC Mahendran Qualcomm Inc. San Diego, California, 92121, USA IEEE International Conference on Communication (ICC 2005).

Speaker: 邱翰銓 Date:2010/10/4

  • Upload
    milek

  • View
    124

  • Download
    0

Embed Size (px)

DESCRIPTION

Using sigcomp to compress SIP/SDP Messages Haipeng Jin and AC Mahendran Qualcomm Inc. San Diego, California, 92121, USA IEEE International Conference on Communication (ICC 2005). Speaker: 邱翰銓 Date:2010/10/4. Outline:. Introduction Signaling compression - PowerPoint PPT Presentation

Citation preview

Page 1: Speaker: 邱翰銓 Date:2010/10/4

Speaker:邱翰銓Date:2010/10/4

Using sigcomp to compress SIP/SDP Messages Haipeng Jin and AC Mahendran

Qualcomm Inc. San Diego, California, 92121, USA

IEEE International Conference on Communication (ICC 2005).

Page 2: Speaker: 邱翰銓 Date:2010/10/4

Outline:

IntroductionSignaling compressionOverview of Text Compression

AlgorithmsPerformance ResultsConclusion

2

Page 3: Speaker: 邱翰銓 Date:2010/10/4

Introduction

SIP and SDP are used in 3G cellular IP MultimediaSubsystems (IMS) to perform multimedia session setup and maintenance. However, both protocols are text-based and have large messages with sizes frequently exceeding several hundred bytes. SigComp can be used to compress these messages and greatly reduce the call setup and feature invocation delays.

3

Page 4: Speaker: 邱翰銓 Date:2010/10/4

Signaling Compression(sigcomp)SigComp is offered to applications such as SIP as a layer between the application and an underlying transport. The service provided is that of the underlying transport plus compression. Basic SigCompDictionaries for SIP/SDP compressionExtended operations

4

Page 5: Speaker: 邱翰銓 Date:2010/10/4

5

Page 6: Speaker: 邱翰銓 Date:2010/10/4

Basic SigCompThe core feature of SigComp is Universal

Decompressor Virtual Machine (UDVM).

The UDVM provides decompression functionalities.

The UDVM is used to decompress SigComp messages.

6

Page 7: Speaker: 邱翰銓 Date:2010/10/4

Dictionaries for SIP/SDP compressionIn order to reduce the session setup time as much as possible, the SIPPING Working Group in IETF introduced a static dictionary for SIP and SDP. The dictionary is to be used in conjunction with SIP, SDP and SigComp.The static SIP/SDP dictionary is a collection of well

known strings that appear in most of the SIP and SDP messages.

7

Page 8: Speaker: 邱翰銓 Date:2010/10/4

Extended operationsExtended operations make use of UDVM instructions and the feedback mechanisms offered by SigComp to significantly improve the compression efficiency .

Dynamic compressionShared compression

8

Page 9: Speaker: 邱翰銓 Date:2010/10/4

Dynamic compressionDuring dynamic compression, a compressor performs compression relative to messages that are sent to the remote end-point prior to the current message.

Shared compression involves compressing the current outgoing message relative to messages received from the remote end-point.

Shared compression

9

Page 10: Speaker: 邱翰銓 Date:2010/10/4

10

1

1

2

9 3

4

56

7

Sigcomp message Requesting feedback

Sigcomp message Returning feedback

8

1

Page 11: Speaker: 邱翰銓 Date:2010/10/4

Overview of Text Compression Algorithms

LZ family of algorithms J.Ziv and A.Lempel

BWT Burrow-Wheeler Transform

PPM Prediction by Partial Math

11

Page 12: Speaker: 邱翰銓 Date:2010/10/4

Dictionary-basedLZ family of algorithmsLZ77LZW

BWT(Burrow-Wheeler Transform)RLE(Run-Length Encoding )

Transform-based

12

Page 13: Speaker: 邱翰銓 Date:2010/10/4

字典索引 索引之對應編碼值 字典內容 1 00 A

2 01 B

3 10 AB

4 11 BB

A B B B A B A A B 1 0 1 1 1 0 00 1 0

Static Dictionary codiog

13

Page 14: Speaker: 邱翰銓 Date:2010/10/4

Model-basedPPM (Prediction by Partial Math) Prediction by partial matching (PPM) is an adaptive statistical data compression technique based on context modeling and prediction.Huffman CodingArithmetic Coding

14

Page 15: Speaker: 邱翰銓 Date:2010/10/4

A B C D E 15 7 6 6 5

A 15

B 7

C 6

D 6

E 5

0

1

11

0

1

13 0

1

24

0

1

39

A B C D E

01

01

0011

11

11

0

Huffman coding

15

Page 16: Speaker: 邱翰銓 Date:2010/10/4

Arithmetic code這個編碼的方法是將訊息全部壓縮成一個,範圍在[0 1﹐ )中由裡面的某一個數來代表這整段訊息說例﹕假設符號  {00 01 10 11}﹑ ﹑ ﹑ 機率分別為 {0.1﹑ 0.2﹑ 0.3﹑ 0.4},依據這些機率分成個子隔﹕[0 0.1)  [0.1 0.3) ﹐ ﹑ ﹐ ﹑       [0.3  0.6) [0.6  1)﹐ ﹑ ﹐  如果二進位訊息輸入為﹕ 10 01 11 00

步驟 輸入符號 編碼間隔1 10 [0.3, 0.6)

2 01 [0.3, 0.33)

3 11 [0.318, 0.33)

4 00 [0.318,0.3192)

16

Page 17: Speaker: 邱翰銓 Date:2010/10/4

0

0.1

0.3

0.6

1 0.33

0.3

0.33

0.318 0.318

0.3192

從 [0.318,0.3192〉中選擇一個數代表輸出:0.3185,所以可以知道 0.3185就是代表的編碼過後的訊息。

17

Page 18: Speaker: 邱翰銓 Date:2010/10/4

Performance Results

18

Page 19: Speaker: 邱翰銓 Date:2010/10/4

Compression Performance for ConcatenatedSIP Messages, size = 7161Bytes

Program CompressedSize (Bytes)

Ratio(%)

Description

zip 913 12.75 LZ77/Deflate

gzip 785 10.96 LZ77/Deflate

compress 2813 39.28 LZW variant.

ppmd 719 10.04 PPM variation

pzip 651 9.09 PPMZ

bzip2 992 13.85 BWT variant.

szip 930 12.99 Related to BWT.

19

Page 20: Speaker: 邱翰銓 Date:2010/10/4

The following notations are used to indicate what states are used in the compression:

SD: static dictionary

SUD: both static and user dictionary

DC: dynamic compression

SC: shared compression

SDC: both dynamic and shared compression

20

Page 21: Speaker: 邱翰銓 Date:2010/10/4

M: message with the maximum size

L: latest sent or received message

ML: both the message with the maximum size and the latest message

We use the following notations toindicate what messages are used:

21

Page 22: Speaker: 邱翰銓 Date:2010/10/4

LZ77 DeflateSize Ratio Size Ratio Diff

No Dictionary 7407 103.4 4413 61.63 41.81

SD_ONLY 3938 54.99 2916 40.72 14.27

SUD_ONLY 2874 40.13 2250 31.42 8.71

SD_DC 1542 21.53 1250 17.46 4.08

SD_SC 1211 16.91 1005 14.03 2.88

SD_SDC 998 13.94 859 12.00 1.94

SD_M_DC 1686 23.54 1354 18.91 4.64

SD_M_SC 1348 18.82 1101 15.37 3.45

SD_M_SDC 1051 14.68 899 12.55 2.12

SD_L_DC 1795 25.07 1427 19.93 5.14

SD_L_SC 1396 19.49 1130 15.78 3.71

SD_L_SDC 1066 14.89 903 12.61 2.28

SD_ML_DC 1644 23.24 1331 18.59 4.65

SD_ML_SC 1273 17.78 1042 14.55 3.23

SD_ML_SDC 1011 14.12 873 12.19 1.9322

Page 23: Speaker: 邱翰銓 Date:2010/10/4

SD SUDSize Ratio Size Ratio Diff

DICT_ONLY 2916 40.72 2250 31.42 9.30

DC 1250 17.46 1038 14.50 2.96

SC 1005 14.03 938 13.10 0.94

SDC 859 12.00 793 11.07 0.92

M_DC 7354 18.91 1111 15.51 3.39

M_SC 1101 15.37 1016 14.19 1.19

M_SDC 899 12.55 826 11.53 1.02

L_DC 1427 19.93 1182 16.51 3.42

L_SC 1130 15.78 1049 14.65 1.13

L_SDC 903 12.61 821 11.46 1.15

ML_DC 1331 18.59 1092 15.25 3.34

ML_SC 1042 14.55 965 13.48 1.08

ML_SDC 873 12.19 804 11.23 0.9623

Page 24: Speaker: 邱翰銓 Date:2010/10/4

Effects of User Defined Dictionary onIndividual Messages (Deflate with DC)

24

Page 25: Speaker: 邱翰銓 Date:2010/10/4

SigComp for IMS Call Setup Messages

25

Page 26: Speaker: 邱翰銓 Date:2010/10/4

ConclusionIt is shown that dictionary-based compression

schemes are the ones best suited for SigComp because of their low complexity, fast speed and low memory requirement.

Static and user-defined dictionaries are very useful for compressing the first several messages in a SIP dialog.

26