CPSC538A Paper presentation 1
Paper Summary:
This paper addressed the question that whether the large-scale live streaming application end-points was feasible based on the current network overlays. In order to answer this question, they first have done a study on the workload of the current network---Akamai. They logged the stream data from Akamai's edge nodes for three months. They classified the stream logs into three catalogs: audio stream(<80kbps), video stream(>80kbps).
After that, they presented three key requirements to answer the feasibility question. (i) Is there enough resource for the overlay? They used four steps to statistic the 90% of the available outgoing bandwidth. For the rest 10% of unknown bandwidth, they used a reasonable measurement. They also analyzed the different resource situation based on the single tree and multiple-tree. All these two situation illustrated that there were abundant resources for the video streams. (ii) Is the network stable with the dynamic change of the participant group? When a node leave the network, how to deal with it's children's connection? They presented the stability metrics for the Overly Protocol and Parent Selection Algorithms to handle this problem. (iii) Could the overlay been constructed efficiently? In this section, they developed the application into clusters.
By all the above three questions' analysis, They came up with the conclusion that it was promising for using such architectures for real-world application end-points.
Discussion Summary:
Three month log analysis is not enough.
The video stream is really low based on the Akamai overlay.
The multiple tree protocol hasn't been illustrated very clear in this paper.
This paper provided good benchmarks for estimating the feasibility for live-stream application end-point.