IMRA 2.0 is an enhanced version of a marathon application for outdoor running events designed to be deployed in 5G infrastructures. During such events, it is impossible for the viewers to monitor the entire race including their athlete/s of interest. IMRA (Immersive Marathon Runner App) is a crowdsourcing application where users can stream video content from the race, from their point of view. At the same time, all the available content is being processed in the FLAME platform to provide back to the users personalized content regarding the athlete that intrigues them. This solution hopes to provide to the viewers of running events with a personalized, based on their interests, coverage of the event, while providing high-level and sophisticated content that derives from Deep Learning and Machine Learning methodologies.
Experiment’s aims:
- Performing module and deployment optimizations such as (i) adapting IMRA 2.0’s modules in-realtime according to the number of requests, slightly compromising quality to maintain low latency; (ii) optimizing IMRA 2.0’s data flow to serve multiple requests at the same time.
- Experimenting with different design patterns to improve the IMRA 2.0 design in terms of QoE and QoS (e.g. smart content placement based on predicted demand for reduction in start-up time for playout. etc.)
- Comparing FLAME’s streaming performance against a cloud-based solution
Key results
During OC4, D-Cube was experimenting with the Geographical Scaling and the Content Placement design patterns. During the remote testing on Bristol platform, D-Cube deployed IMRA modules on all 5 towers, experimented on both scenarios and collect measurements that revealed the gain in QoE.
Geographical scaling
This feature ensures that the deployed resources are utilized effectively and efficiently and that the user will be served by the most-closest endpoint. When the closest endpoint was activated during geographical scaling out, almost x2.5 times lower initial playout delay was achieved.
Content Placement based on prediction demand IMRA 2.0’s Smart Edge Storage module was notified by the Smart Summarization for newer versions of an athlete’s summary and decided when to cache it locally based on previous requests. That way, in case of a future request, content is streamed from the closest node, resulting in lower initial playout time and higher bitrate.
Flame benefits over Cloud
During experimentation, D-Cube also realized the following benefits of streaming from FLAME vs a cloud solution:
- 3.5 times lower latency
- 90% higher bitrate
- Higher user acceptance, as sensitive data always remain local
Next Steps
Experimenting with the deployment of cutting-edge AI services in a 5G platform with distributed resources at the edge allowed to gain insights for the challenges that D-Cube products would face and how to address them, upon 5G availability.
The derived know-how from this open call is of high value and will be used to enrich D-cube’s flagship product “the Immersive Framework”, in order to support 5G topologies for immersive applications.
