Past Research Projects

Below is a summary of some of the past research projects I have worked on.

QoE for Streaming Video

Every year, video traffic makes up a larger percentage of total Internet traffic. Yet we know little about how network conditions affect the user-perceived quality of Internet video, or even how to define or measure user-perceived quality. In order to optimally design the video-based applications of the future, and the application and network protocols to support these applications, we need to be able to measure subjective video quality, from the users' perspective, in a scalable and useful fashion.

In this project, we identified application-layer measurements that can be used to discern and/or predict the user-perceived quality of streaming video. User-perceived quality indicates how a user would rate the subjective quality of the video compared to videos s/he has watched in the past. Streaming video means video retrieved on-demand from a media server and played out as it is received at the media client. Application-layer measurements are used because of their proximity to the user, their ability to reflect network conditions, and their relative ease of collection and analysis. We instrumented a version of Windows Media Player to collect and log stream state information (packet arrival rates, average bandwidth, current frame rate, etc.) periodically, as well as collect subjective user quality ratings (first on a 7-point scale, then on a continuum). We apply data mining techniques to the collected data to assign quality ratings to video streams. Using this approach, we identified several measurements that are excellent indicators of user-perceived video quality---our data mining techniques can assign correct ratings to past streams with 70-90% accuracy.

We expanded this work into RTMP and HTML5 video streams, developing a Flash plugin to collect and analyze data from Flash streams. Our results indicate that 70-80% accuracy in predicting video QoE, using as little as 20 seconds of stream state information, is attainable.

This work has been sponsored by grants from the Howard Hughes Medical Foundation and Carleton College. Earlier versions of this work were sponsored by Hewlett-Packard Laboratories.

Project publications

  1. A. Csizmar Dalal. "User-perceived Quality Assessment of Streaming Media Using Reduced Feature Sets". In ACM Transactions on Internet Technology 11(2), December 2011.
  2. A. Csizmar Dalal. "Systems Considerations in Real Time Video QoE Assessment". In Proceedings of the IEEE Workshop on Quality of Experience for Multimedia Communications (QoEMC), Anaheim, CA, December 2012.
  3. A. Csizmar Dalal, Andy Bouchard, Sara Cantor, Yiran Guo, Anya Johnson. "Assessing QoE of On-Demand TCP Video Streams in Real Time." In Proceedings of the IEEE International Conference on Communications (ICC 2012), Ottawa, Ontario, Canada, June 2012.
  4. H. French, J. Lin, T. Phan, A. Csizmar Dalal. "Real Time Video QoE Analysis of RTMP Streams." (Poster). In Proceedings of the 30th IEEE International Performance Computing and Communications Conference (IPCCC), Orlando, FL, November 2011.
  5. A. Csizmar Dalal. "Revisiting a QoE Assessment Architecture Six Years Later: Lessons Learned and Remaining Challenges". (Invited paper) In Proceedings of the Third International Workshop on Advanced Architectures and Algorithms for Internet DElivery and Applications (AAA-IDEA), Las Palmas de Gran Canaria, Spain, November 2009.
  6. A. Csizmar Dalal, E. Kawaler, S. Tucker. "Towards Real-Time Stream Quality Prediction: Predicting Video Stream Quality from Partial Stream Information". In Proceedings of The Sixth International ICST Conference on Heterogeneous Networking for Quality, Reliability, Security and Robustness (QShine), Las Palmas de Gran Canaria, Spain, November 2009.
  7. A. Csizmar Dalal and J. Olson. "Feature Selection for Prediction of User-Perceived Streaming Media Quality." In Proceedings of the 2007 International Symposium on Performance Evaluation of Computer and Telecommunication Systems (SPECTS), San Diego, California, July 2007. (pdf)
  8. A. Csizmar Dalal, D. Musicant, J. Olson, B. McMenamy, S. Benzaid, B. Kazez, E. Bolan. "Predicting User-Perceived Quality Ratings from Streaming Media Data". In Proceedings of the IEEE International Conference on Communications (ICC 2007), Glasgow, Scotland, June 2007. (pdf)
  9. A. Csizmar Dalal and K. Purrington, "Discerning User-Perceived Media Stream Quality Through Application-Layer Measurements." In Proceedings of the First International Conference on Multimedia Services Access Networks, Orlando, Florida, June 2005. (pdf)
  10. A. Csizmar Dalal and E. Perry. "A New Architecture for Measuring and Assessing Streaming Media Quality." In Proceedings of the Third Workshop on Passive and Active Measurements (PAM 2003), San Diego, CA, April 2003. (pdf).

Queueing Analysis of Application Behavior

(PhD dissertation, Northwestern University)

My dissertation research focused on using queueing theory to analyze networked application performance. My main work focused on web servers and was later extended to more general telecommunication systems (cellular call handoffs, overload control in call processing systems, transmission of real-time packets over a packet-switched network).

Web users are notoriously impatient, willing to wait only a few seconds before aborting a pending web request. This behavior unnecessarily taxes web server resources, which are dedicated to a connection that ultimately does not complete. While this is not as problematic at well-provisioned or lightly-trafficked web servers, where all incoming requests can be serviced with little or no queuing time required, such behavior is extremely detrimental to poorly-provisioned or very busy web servers, where server deadlock may occur and no users are adequately served.

In this work, I analytically derived, using queuing models, a service ordering for web servers that is useful under these extreme conditions and that takes into account the impatience of web users. The ordering is based on a revenue generation model that assumes a web server earns some amount from processing a user request to completion. This amount diminishes exponentially with the time spent in queue and in service at the web server. In such cases, a greedy algorithm maximizes server revenue and outperforms "fair" algorithms such a first-in-first-out by several orders of magnitude when server load is high. These performance gains continue even when the server passes into an overload situation, allowing the web server to operate even when completely saturated. These results are verified both analytically and via simulation, and have been shown to be useful for general Markovian queuing systems as well.

Project publications

  1. A. Csizmar Dalal and S. Jordan, "Optimal Scheduling in a Queue with Differentiated Impatient Users." In Performance Evaluation 59(1), January 2005, pp. 73-84. (pdf)
  2. A. Csizmar Dalal and S. Jordan. "An Optimal Service Ordering for a World Wide Web Server." Performance Evaluation Review 29(2), September 2001, pp. 8-13. (pdf)
  3. A. Csizmar Dalal and S. Jordan. "Improving User-Perceived Performance at a World Wide Web Server." In Proceedings of Globecom 2001, San Antonio, Texas, November 2001. (pdf)
  4. A. Csizmar Dalal. "Characterization of User and Server Behavior in Web-based Networks." Ph.D. Dissertation, Northwestern University, December 1999. (pdf)

Source Characterization of H.323 Traffic

In this joint project with Ikhlaq Sidhu and Jerry Mahler at 3Com, I measured and characterized videoconferencing traffic on a LAN using Microsoft's NetMeeting software. I determined the individual load placed on the network due to audio traffic and used this to deduce the load due to video and "collaboration" (document sharing) traffic. I calculated the statistics and derived distributions of packet sizes, packet interarrival times, packet arrival rates, and byte arrival rates for each of the traffic types and for the three phases of the session: set-up, data transfer, and tear-down. Additionally, I determined the mix of UDP and TCP packets for each phase of the session. These results formed the basis for later telepresence work at 3Com.

Project publications

  1. A. Csizmar, I. Sidhu, J. Mahler. "Source Characterization of H.323 Multimedia Traffic." Technical report, Advanced Technologies Research Center, Carrier Systems Business Unit, 3Com, February 1998.

Analysis of a Cellular Digital Packet Data (CDPD) Network

(M.S. project, Northwestern University and Ameritech Cellular Services)

Cellular Digital Packet Data, or CDPD, provides a way to transmit packetized data over a non-circuit-switched connection in an analog cellular network. Because data traffic shares the cellular channel with voice traffic, where voice is given preemptive priority, the reliability of the data transfer channel is a concern. Working jointly with Ameritech Cellular Services, I developed a simulation model to determine how and when channel congestion occurs, in terms of blocked data and/or voice calls, and how the presence of data traffic affects the quality of voice traffic and vice versa in both dedicated-channel and channel-hopping networks. The simulations showed that for a mixed population of mobile voice and data users, channel-hopping networks had lower blocking and outage probabilities for voice traffic and acceptable throughput and blocking/outage probabilities for data traffic than dedicated-channel networks.

Project publications

  1. A. Csizmar. "Performance of a Cellular Digital Packet Data Network." M.S. Project Report, Northwestern University, December 1997.
  2. A. Csizmar. "CDPD Security Issues." White paper, Ameritech Cellular Services, June 1996.