Resource consumption-aware QoS in cluster-based VOD servers☆
Introduction
VOD is a representative streaming media service technology and has been researched in various areas. However, it is hard to expect good performance if the VOD system is implemented without considering the interrelation between the server and client [1]. The performance of the VOD system is represented with the number of concurrent clients giving a guarantee against a stable QoS. The QoS is closely related to the server architecture model, storage systems, network capability and client terminal types.
In this paper, the cluster-based VOD server denoted as the VODCA (Video-On-Demand Cluster Architecture) is presented. Our devised system is composed of a general PC cluster platform and performs parallel processing for MPEG 1, 2 media. Not only does the VODCA provide the scalable performance within the pre-defined QoS ranges, but also it supports a various requests from clients including VCR functions.
For the parallel processing of MPEG 1, 2 media, the granularity for parallelized materials should be determined. In our research, the Group of Pictures (GOP) layer in MPEG is regarded as a granularity unit. The MPEG movie data is striped as GOP units and evenly distributed to backend nodes. The GOP units extracted from MPEG media are independently processed in backend nodes. Based on parallel storage, parallel retrieval and parallel transmission into a network, the load balancing is maintained among backend nodes. From our approaches, the VODCA can supply the scalable streams to clients within the pre-defined QoS metrics.
Due to intrinsic MPEG characteristics, it is hard to implement VCR functions such as fast rewind, fast forward and resume commands in VOD servers [2], [3]. If these functions are implemented by means of sending and playing the movie data at a faster speed, the network is easily saturated. In our study, we propose VCR functions by extracting I frames from MPEG data and by managing these frames independently. Our experiment shows that the proposed method for VCR functions reduces the network traffic by sending only I frames.
The PC is usually considered as the standard equipment of VOD clients, but we employ TV to provide non-specialists with a friendlier interface. To satisfy the interactive requirements of clients on the TV side, we implement a set-top box that is composed of an embedded board and an infrared sensor device controlled by a remote controller. This TV client provides a more comfortable human interface by selecting desired movies within the range of the remote controller.
In the VOD system, admission control is required to guarantee the quality of streaming media while all clients are being serviced. In our research, we devise system monitoring functions to analyze resource consumption for serviced streaming media. Based on these functions, we measure the quantity of memory, disk bandwidth and network bandwidth consumed for satisfying various client requests with QoS streaming movies. From these detailed measurements, the bottlenecks of scalable performance in cluster-based VOD servers are investigated. Based on these analyses of resource consumption, a RCAAC (Resource Consumption Aware Admission Control) method is proposed. It is based on available system resources and the amount of resources consumed for QoS streams. In our experiments, we confirm that the proposed method increases the utilization of the system resource by guaranteeing the maximum number of QoS streams.
The rest of this paper is organized as follows. Section 2 discusses the parallel processing of MPEG data and the implementation of VCR functions. Section 3 provides the details for our VODCA system. In Section 4, the performance of our VOD system is measured and the results are analyzed. Section 5 makes suggestions for the resource consumption-aware admission control. Section 6 describes related work and Section 7 concludes the paper.
Section snippets
Characteristics of MPEG media
MPEG-1, 2 media consists of a video sequence layer, a Group of Pictures (GOP) layer and a picture layer. The video sequence layer is a group of sequential pictures that have the same frame size and frame rate. The GOP layer is a minimal unit for playing movies and is usually exploited at the random access unit. The picture layer is the single image displayed on the screen. This picture layer is composed of 4 kinds of frames known as I, B, P, D frames. Each frame has its own different function
Architecture of the VODCA system
To examine out performance limitations in large scale VOD services, we implemented the VODCA (Video-On-Demand on Clustering Architecture) system. The VODCA system includes not only server sides but also client sides. Servers in the VODCA consist of a HS (Head-end Server) node and several MMS (Media Management Server) nodes known as backend nodes. The client system in the VODCA is working together with HS and MMS nodes. Fig. 2 shows the architecture of our VODCA system.
Experiment environment
The VODCA server for our experiments consists of a HS node and 6 MMS nodes. Each node operates on the Linux operating system. The MMS nodes, HS node and clients are connected via a 100 Mbps Ethernet switch. All applications included the system administrative tools of the HS node are developed on Qt, C and C++ libraries. Table 1 shows the hardware components for each MMS node in the VODCA system.
We use the yardstick program to measure the performance of our cluster-based VOD servers [11]. The
Resource consumption-aware admission control
From the last section, the performance of the VODCA system is studied on the limited internal resources. To analyze the causes of performance degradation, we measured the amount of resources consumed for each QoS stream by exploiting the monitoring functions. Within the implemented VODCA system, it was clear that the main memory caused the major bottleneck and the network bandwidth did not affect the scalable performance greatly. However, to get the scalable performance, there were no burdens
Relate work
Many research were undertaken for VOD systems to provide a stable service to more users under the various user requirement features and limited resources [13], [14], [16], [19], [20], [21], [23]. Many VOD systems were implemented and studied for research purposes or for commercial purposes [13], [16], [17], [18], [23]. There were several problems in the previous VOD services. They did not support VCR like functions and actual admission control mechanism in implemented levels because the
Conclusion
For successful VOD systems, it is important to provide QoS streams to more clients and a more user friendly interface. In this paper, we studied the performance issues for supporting QoS streams in cluster-based VOD servers. Based on our implemented VODCA system, the performance scalability is investigated according to the number of MMS nodes and various client demands including VCR functions are supported. Experiments have shown that a nonlinear scalability phenomenon occurred over the
Dongmahn Seo received his B.E. and M.E. degrees in Computer Engineering and Computer Information and Telecommunication Engineering from Kangwon National University, in 2002 and 2004, respectively. He is currently a Ph.D candidate in Computer Engineering at Kangwon National University. His research interests include multimedia system, parallel processing, embedded system and wireless sensor network.
References (29)
- et al.
Multimedia Servers: Applications, Environments, and Design
(2000) - W.C. Feng, M. Lie, Critical bandwidth allocation techniques for stored video delivery across best-effort networks, in:...
- D.H.C. Du, Y.J. Lee, Scalable server and storage architectures for video streaming, in: IEEE International Conference...
- et al.
Techniques and Standards for Image, Video, and Audio Coding
(1996) - ...
- Jung-Min Choi, Seung-Won Lee, Ki-Dong Chung, A muticast delivery scheme for VCR operations in a large VOD system, in:...
- Gerard Beekmans, Linux From Scratch Version 3.3. Available from:...
- Tome Fawcett, The Linux Bootdisk HOWTO. Available from:...
- Qt/Embedded Whitepaper. Available from:...
- ...
Multimedia storage servers: a tutorial
IEEE computer
Integrated system for multimedia delivery over broadband ip networks
IEEE Transactions on Consumer Electronics
Cited by (6)
Grouping and time-series notifying of periodic data in a real-time streaming system for smart toy claw machine
2019, Journal of Systems ArchitectureCitation Excerpt :An encoding/decoding computation based on the available network bandwidth was proposed for real-time streaming [19,20]. A grouping configuration of cluster nodes for Video-On-Demand (VOD) systems was proposed to provide Quality of Service (QoS) to more clients under limited resources [21]. A low-complexity algorithm based on background subtraction and error resilience techniques was proposed to reduce the transmission bandwidth of a video stream [22].
Network-adaptive autonomic transcoding algorithm for seamless streaming media service of mobile clients
2011, Multimedia Tools and ApplicationsLoad distribution algorithm based on transcoding time estimation for distributed transcoding servers
2010, 2010 International Conference on Information Science and Applications, ICISA 2010Parallel failure recovery techniques in cluster-based media servers
2010, Journal of SupercomputingRecovery Strategies for Streaming Media Service in a Cluster-Based VOD Server with a Fault Node
2009, International Journal of Parallel ProgrammingParallel failure recovery techniques in cluster-based media servers
2008, Proceedings of the 2008 International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA 2008
Dongmahn Seo received his B.E. and M.E. degrees in Computer Engineering and Computer Information and Telecommunication Engineering from Kangwon National University, in 2002 and 2004, respectively. He is currently a Ph.D candidate in Computer Engineering at Kangwon National University. His research interests include multimedia system, parallel processing, embedded system and wireless sensor network.
Joahyoung Lee received his B.E. and M.E. degrees in Information and Telecommunication Engineering and Computer Information and Telecommunication Engineering from Kangwon National University, in 2003 and 2005, respectively. He is currently a Ph.D candidate in Computer Engineering at Kangwon National University. His research interests include multimedia system, parallel processing, embedded system and wireless sensor network.
Yoon Kim received his B.S., M.S., and Ph.D. degrees in Electronic Engineering from the Department of Electronic Engineering, at Korea University, in 1993, 1995, and 2003, respectively. From 1995 to 1999, he was with the LG-Philips LCD Co. where he was involved in research and development on digital image equipments. In 2004, he joined the Department of Electrical and Computer Engineering at Kangwon National University where he is currently an Assistant Professor. His research interests are in the areas of video signal processing, multimedia communications, and sensor network.
Chang Yeol Choi received his B.E. and M.E. degrees from Kyungpook National University, and the Ph.D. degree in computer engineering from Seoul National University. He was with ETRI as a principal engineer responsible for a computer system development from 1984 to 1996. His major interests are computer system architecture, multimedia systems, and mobile computing.
Manbae Kim received his B.S. degree from Hanyang University, in 1983 and the M.S. and Ph.D. degrees in Electrical Engineering from University of Washington, Seattle in 1986 and 1991, respectively. From 1991 to 1998, he was with Samsung Advanced Institute of Technology (SAIT), Korea. He is currently associate professor at Kangwon National University. His research interests include MPEG-21, realistic broadcasting system design and multi-view video processing.
Inbum Jung received his B.S. degree from Korea University, in 1985 and the M.S. and Ph.D. degrees in Computer Science from KAIST, in 1994 and 2000, respectively. From 1984 to 1995, he was with Samsung Electronics Co. Ltd., Korea. He is currently a faculty member at Kangwon National University. His research interests include operating system, parallel processing, streaming media and wireless sensor network.
- ☆
This research was supported by the MIC (Ministry of Information and Communication), Korea, under the ITRC (Information Technology Research Center) support program supervised by the IITA (Institute of Information Technology Assessment) (IITA-2005-(C1090-0502-0022)). This work was partially supported by the Kangwon Institute of Telecommunications and Basic Research Program of the Korea Science Engineering Foundation (R05-2003-000-12146-0).