Chapter 7 End-to-End Data Ohanes Dadian, Danny Luong, Yu Lu Contents 7.1 Presentation Formatting 7.1.1 Taxonomy 7.1.2 Examples 7.1.3 Markup Languages 7.2 Multimedia Data 7.2.1 Lossless Compression Techniques 7.2.2 Image Representation and Compression 7.2.3 Video Compression 7.2.4 Audio Compression Definition Data Presentation Format Agreement An important aspect Data manipulation function Encoding What is it? Importance Argument marshalling Decoding Argument unmarshalling 7.1 Presentation Formatting Challenges of Presentation Formatting Different representation in Floating-Point Numbers Different registers sizes IEEE standard 754 format Nonstandard format 16-bit register 32-bit register 64-bit register Different representation in Integers Big-Endian Little-Endian 7.1.1 Taxonomy Data Types Conversion Strategy Tags Stubs Data Types Lowest Level - Base Types Medium Level – Flat Types Integers Floating-point numbers Characters Structures Arrays Highest Level – Complex Types Those types that are built using pointers Data Types Conversion Strategy Canonical intermediate Form Sender: internal representation external representation Receiver: external representation internal representation Receiver-Makes-Right Sender : Do not convert, send its internal representation directly receiver : Any representation internal representation Combined Conversion Strategy Debate Canonical intermediate Form vs. Receiver-Makes-Right 7.2.1 Lossless Compression Techniques Data Compression Data Compression Categories What is it? Bandwidth vs Throughput When to compress? Lossy vs Lossless Lossless Compression Techniques Run Length Encoding Differential Pulse Code Modulation Dictionary-Based Methods Tags Tagged Data A tag = additional information Help the receiver decoding data Untagged Data How does the receiver know how to decode? Think any object-oriented language as an example Stubs On the Client Side Marshal the arguments into a message On the Server Side Converts the message back into variables Stubs 7.1.2 Examples External Data Representation (XDR) Supports the entire C-type System without function pointers defines canonical intermediate form Does not use tags Uses compiled stubs 7.1.2 Examples Abstract Syntax notation One (ASN.1) Supports the C-type system without function pointers Defines canonical intermediate form Uses type tags Uses either interpreted or compiled stubs Representation <tag , length , value> 7.1.2 Examples Network Data Representation (NDR) Supports the C-type system Defines receiver-makes-right Uses tags Generates the necessary stubs Representation 7.1.3 Markup Languages(XML) Extensible Markup Language – XML XML Schema Document – XSD What is xml? Can be sent as data over internet Can be configuration file used in frameworks Defines XML Namespace Solve name clashes URI – Uniform Resource Identifier (detail in Chapter 9) xmlns: emp = “http://www.example.com/employee” <emp: title> head Bottle Washer </emp:title> XML--Example XSD--Example Data Compression Data Compression: Less data for the same message, increase throughput Bandwidth vs Throughput Bandwidth – Physical Throughput – Logical When to compress? Time Cost Data Compression Categories Two different categories Lossy Lossless Pros and Cons of each technique Compression ratio Exact reconstruction Lossless Lossy 7.2.1 Lossless Compression Techniques Data Compression Data Compression Categories What is it? Bandwidth vs Throughput When to compress? Lossy vs Lossless Lossless Compression Techniques Run Length Encoding Differential Pulse Code Modulation Dictionary-Based Methods Data Compression Data Compression: Less data for the same message, increase throughput Bandwidth vs Throughput Bandwidth – Physical Throughput – Logical When to compress? Time Cost Data Compression Categories Two different categories Lossy Lossless Pros and Cons of each technique Compression ratio Exact reconstruction Lossless Lossy Lossless Compression Techniques Run Length Encoding Differential Pulse Code Modulation Dictionary-Based Method Run Length Encoding (RLE) One of the simplest method of data compression A lossless compression method Consecutive data elements saved as one element and a count Simplified Example AAAAAAAAAABBBBBOOOOOOOOO 10A5B9O Real world deals with binary instead Most useful on data that contains repeated elements Differential Pulse Code Modulation (DPCM) Another simple lossless compression algorithm Example Output a reference symbol Output the difference between new symbol and the reference AAABBCDDDD A0001123333 Work better on digital images Dictionary-Based Method The last lossless compression algorithm we discuss The idea is to build a dictionary (common phrases) Replace common phrases with index Dictionary constructed during compression sent along with the code A lot of research on how to build an efficient dictionary 7.2.2 Image Representation and Compression Focus on GIF & JPEG Differences between GIF and JPEG JPEG Compression Phases DCT Phase Quantization Phase Encoding Phase Difference between GIF and JPEG Graphic Interchange Format (GIF) RGB color space 8 bits each dimension = 24 bits total Instead of sending 24 bits, reduce to 8 bits first 28 = 256 color Picture has more than 256 colors, change it and cut it down Good compression ratio, but low quality Difference between GIF and JPEG Joint Photographic Experts Group (JPEG) Most widely used More suited to photographic images Transforming RGB color to YUV color space (lossless) Y component represent brightness (luminance) UV components represent color (chrominance) Human eyes have separate receptors for brightness and color By separating brightness and color, we can perform compression separately Human eyes are less sensitive to color, so we can compress UV component more aggressively (lossy) GIF Compression Phases No phases One single step Lossless Each color is represented by 24 bits Reduce it down to 8 bits Only 256 color possible More than 256 color, merge to closest color Usage Sharp-edged images (logos) JPEG Compression Phases 3 different phases DCT (lossless) Quantization (lossy) Encoding JPEG Compression Phases Discrete Cosine Transform (DCT) Phase Closely related to the Fast Fourier Transform (FFT) 8x8 matrix of pixel value => matrix of frequency coefficients Help Filter out least important pattern Lossless Formula JPEG Compression Phases Quantization Phase DCT phases only transform the data into something easier to recognize Quantization Phase drops the insignificant bits of the frequency Coefficients Truncate information => Lossy Dropping the insignificant bits of the frequency coefficients JPEG Compression Phases Encoding Phase Final phase Encodes the quantized frequency coefficients in a compact form Coefficients are processed in zigzag sequence Further compression can be applied e.g. Run Length Encoding 7.2.3 Video Compression What is a video? Frames can be compressed using DCT A succession of frames, or images. Like JPEG Compression should take into consideration the redundancy between frames. Two successive frames likely have a lot of the same information, depending on the moving object. MPEG The MPEG format provides standards for multimedia compression. Includes compression, decompression, processing and codedrepresentation. MPEG 1, MPEG 2, MPEG 4 Named after the group that invented it. Moving Picture Experts Group Established in 1988 and included experts from ISO and IEC. How MPEG Works Frame Types MPEG compresses frames into three types. I frames Self-dependent. P frames I frames, P frames and B frames. Specifies difference from the previous I frame. B frames Interpolation between the previous and subsequent I or P frames. Frame Types, contd. Frame Types, contd. Frame Types, contd. 8 x 8 Macroblock Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y U U V V U U V V Effectiveness and Performance MPEG video can be compressed with up to a 90:1 ratio. Encoding/Decoding can be performed on software and hardware. Hardware being the most prominent for encoding. Software being the most prominent for decoding. 7.2.4 Transmitting MPEG over a Network MPEG defines video stream. Complex but modular. Three variations. MPEG 1, MPEG 2 and MPEG 4. Focus on Main Profile MPEG 2 stream. Does not dictate how stream is stored. Used by DVD. Nested Structure MPEG Format Transmitting MPEG over a Network, contd. Allows for trading picture quality for bandwidth. Need to packetize stream. Not an issue with TCP. For UDP, streams are broken at selected points. Need to deal with packet loss. Macroblock boundaries. Application Level Framing. Mark frames with drop probability. Need to deal with application latency constraints. MPEG Video Compression Algorithm 7.2.4 Audio Compression MPEG provides audio compression standard. MPEG I/2 Layer III (MP3). Example: CD quality audio at 44.1 KHz sampling rate requires 1.41 Mbps bitrate. 4.32 Mbps with synchronization and error correction overhead. Some compression is in needed for network transfer. MP3 compression solves this issue. MP3 Compression Algorithm Audio Compression, contd. Optimization lies in the number of subbands used and how many bits are allocated for each subband. Governed by psychoacoustic models. Why is this the case? Take the following for example. A male voice requires more bits to low-frequency subbands. A female voice requires more bits to low-frequency subbands. Change in frequency is done by dynamically changing the quantization tables used for each subband. Summary Discussed how application data is encoded in network packets. Presentation Formatting. Compression. Lossy Lossless. Provide high quality media. Q&A Thank You References Peterson, Larry L., and Bruce S. Davie. Computer Networks: A Systems Approach. 5th ed. Amsterdam: Morgan Kaufmann, 2012. Print. Motion Picture Experts Group. Wikipedia. http://en.wikipedia.org/wiki/Mpeg (accessed February 19, 2013). Extra Slides Frame Types, contd. B Frame Macroblock Coordinate c Motion Vector relative to previous reference frame a Motion Vector relative to subsequent reference frame b Delta for each pixel in macroblock d Frame Types, contd. Goal: Find the corresponding reference pixel using the past and future reference frames. coordinate(d ) : (x p , y p ) Past Reference Frame Fp coordinate(b ) : (x f , y f ) Future Reference Frame coordinate(c) : (x, y) Ff Fc (x, y) = Fp (x + x p, y + y p ) + Ff (x + x f , y + y f ) 2 + d (x, y) Other Video Encoding Standards ITU-T has defined the H series of codecs for video, audio, control and multiplexing. H.261, H.263 First and second generation for video encoding. Targeted for lower speeds. Available in 64-kbps increments. H.264 New generation codec. Part of the MPEG-4 standard. One of the most popular codecs for HD video on web and mobile platforms. Used by Blu-ray, iTunes,YouTube, Flash, Silverlight and various HDTV broadcasts.