 |
 |
Event Reports
June 2010
ACM Multimedia Systems Conference 2010
The inaugural MMSys conference was held in Phoenix, AZ this past February and by all measures was a great success. Twenty-nine papers were presented over two days with lively Q&A discussions. In this article, I want to point out some of the technical highlights from select papers that I personally found most interesting and describe some of the future plans that are in the works for the conference.
The use of GPU-based computing in all facets of computing has been steadily on the rise over the last decade, and multimedia systems are particular well-suited for taking advantage of this hardware resource. "Exploring NVIDIA-CUDA for video coding" by A. Colic, H. Kalva, and B. Furht presented in the very first session of the conference explores the speed up gains for different approaches of employing CUDA in video coding. Among the insights provided by the paper is the sensitivity of CUDA-based processing to the use of global memory and to memory organization in general. While fantastic improvements in performance are possible with CUDA, achieving the most out of a GPU still requires a careful organization of inter-thread communication and alignment of data within memory. The reported performance results for fully optimized code is incredibly impressive, but only comes with implementations born of practice and experience with CUDA-based programming. Another trend in computing these days is the increasing reliance on "cloud-computing" for running hosted applications. For multimedia applications that are latency-sensitive, however, the quality of experience will be greatly affected by the performance achieved by the specific resources within the cloud that are used. A cloud computing platform that makes promises about average-case performance may not be able to characterize the performance achieved at any given time. So just how variable are cloud computing resources with respect to latency and load? That is the subject of "Empirical Evaluation of Latency-sensitive Application Performance in the Cloud" by S. Barker and P. Shenoy from the University of Massachusetts, Amherst. This paper describes the results of experiments conducted on Amazon's EC2 to quantify the CPU, disk, and network jitter and throughput fluctuations observed over several days using both micro-benchmarks, a Doom 3 game server, and Apple's Darwin Streaming Server. The overall conclusion of the paper is that performance of latency-sensitive applications hosted in the cloud can suffer substantially from interference due to background load from other applications and virtual machines also installed within the cloud. The work points toward a need for better controls and interfaces between applications and cloud-based resources in order to negotiate these performance requirements.
The last paper that I want to highlight is "Paceline: Latency Management through Adaptive Output" by A. Erbad, M. Narjaran, and C. Krasic. Often multimedia systems struggle to manage and mitigate performance bottlenecks imposed by the hidden costs associated with the operating system. Socket buffering delays and overhead associated with crossing between the user level and kernel level are good examples. While clean slate solutions that provide for application-directed cross-layer optimization may be academically interesting, finding ways to overcome conventional end-to-end latency shortcomings of TCP at the application-level without requiring custom modification of either TCP or the operating system kernel is important for building practical multimedia systems. Paceline provides a user-level library that exposes a rich interface for message framing, multiplexing, connection management, and latency control that operates on top of TCP without modification and the results in the paper document a factor of 3 improvement in median latency and show it to be competitive with more radical TCP alternatives. Instead of keynote speeches, the MMSys organizing committee opted to include a session of four invited papers from specific researchers from a diverse set of viewpoints. Included in this session were papers by Mark Claypool from WPI on latency management in networked games; R. Micheal Young from North Carolina State University on modelling and representing conflict and intent within narrative-driven virtual worlds; Chris Gauthier Dickey from the University of Denver on the convergence of games, multimedia, and real-world sensors; and Marco Mattavelli from EPFL and associates on reconfigurable video coding and the emergence of new video coding standards. It was clear that the authors were able to take advantage of the invited paper opportunity as a platform for speculating more broadly about their subjects than would normally be possible in a traditional paper and the session was a great success that is likely to be repeated in next year's conference. In addition to the conference, program committee members were invited to a planning meeting held before the conference itself to discuss ways in which we as a community should develop the conference in the years to come. In particular, committee members widely recognized the need to develop closer ties and participation with industry. MMSys can and should become a forum at which industry participants are exposed to the advances being made within academia and provided an opportunity to help improve the quality and relevance of that research. Toward that end, we have partnered with Cisco who will act as host of MMSys 2011 in San Jose, CA and are actively reaching out toward other industrial participants. We hope to work with Cisco to experiment with remote attendance of the conference via their WebEx and tele-presence services. Finally, we have included a Dataset Track to next year's conference in order to facilitate and encourage sharing of real-world datasets. Please see the MMSys 2011 CFP at www.mmsys.org from more details. PapersCoding & processing
System performance
Wireless & mobile
Scalability
Streaming
Invited papers
Search & navigation
Tele-immersion & virtual environments
Reports from other ACM EventsArt.on.Wires 2010
Art.on.Wires is a laboratory space in which creative artists and researchers explore technology for live, interactive and mixed-reality spaces. Art.on.Wires was organised by Alexander Eichhorn, a research scientist at Simula Research Laboratories in Oslo, Norway. It was held as a four-day event from May 10-13 2010 at Kanonhallen, an old industry hall in Oslo.
Art.on.Wires has been arranged in spirit of 9 Evenings, the first large-scale collaboration between artists, engineers and scientists organised by Robert Rauschenberg and Billy Klüver in New York in 1966. Our mission was to explore space and interaction in networked mixed-reality environments. A particular focus was put on the integration of Vision, Sound and Motion technologies into a local performance space to enable meaningful remote interactions between artists. The secondary aims were to make contacts with people from different fields and to start dialogue and collaborations between researchers, engineers and performing artists that are supposed to last for longer.
The 50x15m large space at Kanonhallen was equipped with latest sound, light, video and motion capturing technology which came mostly from research facilities at Simula Research Laboratory, the University of Oslo and the Technical University of Ilmenau, Germany. Over 70 international artists, researchers and students from 9 countries around Europe attended Art.on.Wires and helped making it a big success.
Our laboratory days were packed with exciting keynote talks from Mark Coniglio (TroikaRanch) and Atau Tanaka (Newcastle University), workshops and hands-on expericene sessions. Workshop topics included telematic performance, digitally informed music, interactive environments, motion capturing, hardware hacking, circuit bending and tutorials on Isadora, VVVV, Ableton Live, as well as OpenFrameworks. In the evenings we arranged concerts with Alexander Carot (performing live over Internet with two musicians in Germany) and Cenizero. Live electronic music was performed by Jacob Korn and Lars Graugaard together with Aki Asgeirsson and Atau Tanaka. Thanks to the FeM Streaming Team from TU Ilmenau we streamed live video from all events to the Internet. We are particularly happy about the many cross-disciplinary contacts that emerged for all participants from Art.on.Wires. Certainly, the creative yet relaxed atmosphere helped a lot to make Art.on.Wires a great experience for everyone. MMSJ, Volume 16, Issue 3, June 2010
The Multimedia Systems Journal is the first journal publication sponsored by ACM SIGMM and today published exclusively by Springer Verlag. As a service to Records readers, we provide direct links to Springer Verlag's Digital Library for the papers of the latest MMSJ issue. Papers
|
||||||||
|
|
||||||||
|
||||||||
|
|
