Layered streaming in overlay networks has drawn great interests since not only can it accomodate large scales of clients but also it handles client heterogeneities. However, to our knowledge, there's still a lack of overlay construction (i.e. neighbor selection) approach suited for layered streaming, because i) In existing works neighbors are selected only based on their network conditions. However, a neighbor with good network condition may not be able to provide sufficient layers (e.g. a neighbor in the same LAN). ii) Previous works usually select "good" neighbors for the new node, ignoring that the joining of the new node could also be utilized to improve the performance of existing nodes. In this paper, OCals-a two-stage QoS aware overlay construction approach for layered streaming is proposed. The main contribution of OCals is that i) when selecting neighbor, it considers existing nodes' network conditions and their providing layers as a whole; ii) it guarantees the QoS for the new node as well as improves the QoS for existing nodes so that with the joining of new nodes, the performance of the overlay could be consecutively improved; iii) it's easy to implement and low time cost. Experiments demonstrate that compared with two other approaches: SCAMP (a pure random neighbor selection method) and Narada (a QoS aware method), the throughput and average packet delay of the layered streaming on top of the overlay constructed by OCals can be remarkably improved. Besides, the time spent on joining and recovery is very short.
