A scalable off-line MPEG-2 video encoding scheme using a multiprocessor system

Abstract

Video compression plays a central role in a vast number of multimedia applications but its computational requirements overwhelm the capabilities of any present single processor system. In this paper, we explore the use of parallel machines like the Intel Paragon to compress MPEG-2 video sequences. The motivation is to build a production-based compression facility by exploiting the potential power of the available machine. Given a video sequence or a set of sequences, the aim of the parallel encoder is to achieve the maximum possible encoding rate. A collective scheduling scheme for the processors, I/O nodes, and disks is proposed that provides fast I/O, minimizes the idle times of processors, and enables the system to work in a highly balanced fashion. An efficient data layout scheme for storing video frames is also proposed in order for the I/O to sustain the desired data transfer rates. Using a small percentage of processors as the I/O nodes results in an efficient utilization of the system resources. As shown by experimental and analytical results, the encoding scheme is scalable and higher performance can be achieved with larger machines. The performance of the proposed scheme can be many times the real-time encoding rates with Standard Interface Format (SIF) and CCIR-601 video sequences. The experimental results indicate about two-fold gain in performance compared to the previous studies. Such a system is useful for the conversion of analog videos to compressed digital form in large studios, digital libraries, and other multimedia database environments. The proposed scheme partitions the system into groups of compute nodes, and I/O nodes, and can be easily extended to other MIMD machines or a set of networked workstations.

Introduction

The realm of high-performance parallel and distributed computing is expanding beyond the traditional scientific community because the current revolution of information technology has created a vast number of commercial applications that require massive computing power [5]. The need for high-performance computing in commercial applications is being recognized by researchers from the areas of both parallel processing and information technology. Large-scale databases, video servers, visualization tools, content-based search and retrieval, graphics rendering, etc., are typical applications that require large computing power.

Video encoding (also called compression) is another application which requires enormous computing power and thus can benefit from high-performance computing. Digitized video, which is a fundamental component of common multimedia applications, typically needs massive amount of data required to represent the audio-visual¹ information. A few minutes of video sequence requires Gbytes of data storage. Similarly, the amount of data for transmitting a video sequence over a network can easily overwhelm the network channels. For example, to transmit an uncompressed high quality digital television signal (CCIR-601) would require 126 Mbps. Even for a lower resolution signal suitable for video conferencing applications (Common Intermediate Format), the uncompressed bit rate is 36.5 Mbps [8]. These amounts become out of reach for high-definition television (HDTV) [9], which will be using digital video at a much higher resolution of 1920×1152.

Storage and transmission of a huge amount of data inevitably calls for compression and decompression of digital video. Fortunately, digital video contains ample redundancies in both the spatial and temporal domains, enabling encoding algorithms to achieve a high degree of compression with little degradation in quality. The entire compression/decompression process requires a codec consisting of an encoder and a decoder. The encoder compresses the data at the transmission or storage end while the decoder decompresses the data for reproducing the video to be viewed by the user. In order to guarantee exchange of the compressed video data between different systems such that a unique decompression is possible for a particular encoded bitstream regardless of the decoder configuration, video coding algorithms are standardized. Two recent international standards, known as MPEG-1 and MPEG-2, have been developed by the moving pictures expert group (MPEG) of the international organization for standards (ISO). These standards specify the syntax (representation) for the decoding of video and accompanying audio data.

Compression is considerably more complex as compared to decompression (which can possibly be done using a single processor machine). The objectives of an efficient video compression techniques include (assuming a defined bit rate) a good visual quality evaluated through subjective and objective assessment criteria, high compression ratio, and low complexity of the compression algorithm itself — the emphasis on any of these objectives can, of course, vary according to the target applications. Compression can be aimed for real-time or non-real-time encoding environments. In the former case, compression must be achieved on-line, for example in a system in which a video stream is being generated from a source such as a video camera; a real-time encoding rate is about 30 frames/s. In the latter case, compression can be done off-line without strict requirements of real-time compression rate. Non-real-time compression is required in applications like digital library or production systems which require encoding a video sequence and storing it on a CD-ROM or digital versatile disk (DVD).

There are two approaches to performing video compression: hardware-based [1], [23] and software-based [2], [3], [4], [7], [10], [11], [19], [20], [22], [24]. Both approaches have their own advantages and disadvantages. A hardware approach uses a special-purpose architecture, and its advantages include the ease of use and high compression speed. However, dedicated hardware is less flexible and can become obsolete. Furthermore, hardware is often optimized for a particular coding algorithm, and cannot be used for exploring other present and future video compression standards. A software solution using general-purpose computing platforms, on the other hand, is more flexible, and thus allows algorithmic improvements. In addition, for non-real-time applications, a software implementation can produce better quality video compression by tuning various parameters and by allowing multiple passes for optimization. However, the very high computation requirements of video applications can often overwhelm a single-processor sequential computer [2], [21]. Therefore, it is natural to exploit the potentially enormous computing power offered by parallel computing systems.

In this paper, we explore the use of parallel machines like the Intel Paragon to compress MPEG-2 video sequences. The motivation is to build a production-based compression facility by exploiting the potential power of the available machine. In addition, parallel machines which are normally available for scientific computing can be exploited for new multimedia applications. The compression facility is aimed to compress multiple and large video sequences. The environment is not real-time but the aim is to achieve the maximum possible encoding rate beyond the real-time speed. Such a facility is useful for the conversion of analog videos to compressed digital form in large studios, digital libraries, and other multimedia database environments. We propose schemes for data layout, efficient I/O, and load-balanced data distribution. These schemes provide fast data retrieval as well as efficient scheduling and matching of I/O and computation rates such that the entire machine operates in a highly balanced fashion without any bottlenecks. Using a very small percentage of processors as the I/O nodes results in an efficient utilization of the system. More importantly, our scheme is scalable, that is, an increase in the number of processors will result in a proportional increase in the encoding rate. As a result, larger machines will yield higher encoding rates. Specifically, given any MIMD machine configuration (that is, the number of processors, I/O nodes, and disks), our proposed method will logically configure the machine for the best possible utilization and match the I/O and encoding rates to reach the ideal performance level.

The rest of this paper is organized as follows. Section 2 provides an overview of MPEG-2, followed by Sections 3 which briefly discusses the related work and gives a motivation for pursuing this research. Section 4 gives an overview of the Intel Paragon and its logical partitioning used in our encoding scheme. Section 5 includes a discussion of the proposed parallel encoder and various related issues. Section 6 presents the experimental results. Section 7 discusses the scalability of the encoder and Section 8 provides some concluding remarks.