Exploiting Proxy-Based Transcoding to Increase the User Quality of Experience in Networked Applications Maarten Wijnants Patrick Monsieurs Peter Quax Wim Lamotte

Overview  Introduction  Intelligent Proxy Architecture Application and Network Awareness Video Transcoding Plug-in  Sample NVE application  Experimental Results  Conclusions  Future Work  Questions

Introduction  Increasing heterogeneity in end-user device space PC, PDA, smartphone,... each with their own capabilities and constraints  Heterogeneity complicates multimedia content delivery i.e. displaying MM content on PC screen vs PDA screen  Possible solutions Providing specialized versions of content On-the-fly content transcoding

Introduction  Supplying specialized versions of content users receive version that best suits the capabilities of their client device and network connection lacks flexibility complicates content management  On-the-fly content transcoding content is transformed to a convenient format before it reaches the end-user flexible solution only high quality content on server however: computational complex

Introduction  Usability of transcoding is not confined to content delivery scenarios  We discuss the implementation of a novel plug-in that adds transcoding functionality to our previously introduced intelligent proxy  Can our transcoding-enabled proxy positively affect the user Quality of Experience (QoE) in networked applications?

Intelligent Proxy Architecture  Our intelligent proxies are application aware client application is enhanced with a “Network Intelligence Layer” the NILayer continuously queries the application's awareness management model retrieved information about the relative importance of different application data streams is sent to proxy  NILayer is highly reusable enables easy integration of our proxy system in many networked applications

Intelligent Proxy Architecture  Our intelligent proxies are also network aware proxy periodically probes the network links going to connected clients  latency  throughput  packet loss rate proxy records bandwidth usage of every data stream that passes through it

Intelligent Proxy Architecture  Based on their compound awareness, our proxies intelligently manage each client's available bandwidth suppose a client's downstream bandwidth no longer suffices to receive all streams at maximum quality proxy will determine which streams should be reduced in quality or even blocked  We position our proxies close to end-users Stream management complexity is moved to the edge of the network

Video Transcoding Plug-in  Our proxies are very generic by default, the proxy can execute only basic operations on data streams  Proxies are equipped with a plug-in mechanism can be used to extend proxy performance and functionality  This approach ensures that our proxies can be integrated in many networked applications and attain a high level of performance in all these situations

Video Transcoding Plug-in  To add more fine-grained stream control to our proxy architecture, we implemented a video transcoding plug-in plug-in enables the proxy to transcode video streams to lower quality by reducing  video bitrate  spatial or temporal resolution uses the cascaded pixel-domain transcoding approach  not very efficient  best results in terms of image quality

Sample NVE application  We tested our proxy architecture by integrating it into our in-house developed multi-user NVE framework focuses on scalability  by maximizing client responsibilities  by relying on direct client-to-client multicast communication tries to increase the immersive experience by supporting video-based avatars

Sample NVE application  Technical details of the NVE framework virtual world is divided into regions  each region has a unique MC address  event information is sent only to the MC address of the originating region Area of Interest manager selects the regions a user should be aware of  limits the amount of information clients need to receive and process virtual world is also divided into video regions  each video region has 3 MC addresses  video-based avatars send out 3 distinct video qualities (high, medium and low)

Sample NVE application  Technical aspects of the NVE framework video qualities used by the framework like the AoI manager, the Video Area of Interest (VAoI) manager decides which video MC addresses the client should subscribe to High QualityMedium QualityLow Quality CodecH263 ResolutionCIF (352x288) FPS2515 Bitrate

Experimental Results  Integrating our proxy architecture in the NVE framework proved to be easy inserting the NILayer into the client software was straightforward transcoding plug-in had to be tailored to the framework

Experimental Results  Description of the experiment 4 clients running the unmodified framework 2 clients connected through our transcoding- enabled proxy clients C1, C2 and C3 were represented as video-based avatars available downstream bandwidth of clients PA and PB was varied artificially over time all clients remained stationary

Experimental Results  Video network traffic received by client PA

Experimental Results  Video network traffic received by client PB

Experimental Results  Video network traffic received by client C4

Experimental Results  What does this experiment learn us? clients PA and PB both roughly stayed within their bandwidth limitations proxy exploits its application awareness to intelligently decide which video streams should be transcoded to a lower quality if the world is populated solely by clients connected through our proxy system, video-based clients only need to send HQ video  less client processor load  asymmetric nature of most current Internet subscriptions

Experimental Results  Main drawback of the video transcoding plug-in is its computational complexity transcoding performance was not one of our main objectives we nonetheless wanted the number of clients our proxy can support simultaneously to be as high as possible  Transcoding plug-in is equipped with an intelligent scaling mechanism if the proxy's average processor usage exceeds a predefined threshold, MQ transcoding is disabled

Experimental Results  The intelligent scaling mechanism can decrease the proxy’s processor load by a factor of two approximately

Conclusions  We have presented a proxy architecture that is both application and network aware  Our transcoding-enabled proxy at all times exploits all available client bandwidth without ever exceeding it, and intelligently and dynamically distributes this bandwidth over all involved network streams in order to optimize the user QoE  scarce client upstream bandwidth is no longer wasted

Future Work  Add device awareness to our intelligent proxy architecture make more complex transcoding decisions distribute client bandwidth even more efficiently  Increase the performance and scalability of our transcoding plug-in  Add transcoding functionality for data other than video (e.g. audio)  Perform more elaborate experiments

Questions?

Additional Results