| Multimedia Distribution (2) |  | 
 
 
 Let us first consider the case of a video stream addressed to one 
target only.
 
Let us first consider the case of a video stream addressed to one 
target only. 
The path between the source 
  and the target is automatically determined at run-time, by tracing the route 
  via one dummy packet sent from source to target (it can be also predetermined 
  by the VoD source according to some previously collected routing information). 
  QoSNs move to the chosen hosts on the path and interrogate the AC database: 
  if available resources are not enough for the desired QoS, QoSNs can coordinate 
  and reduce their requests by scaling the stream (at 
  the moment, by dropping frames in Motion JPEG streams or by reducing resolution 
  in MPEG-2 ones). Only if these diminished reservation requests cannot be satisfied, 
  the VoD service is denied. After a successful negotiation phase, the (possibly 
  scaled) multimedia stream starts to flow.
During 
  the video distribution, a link can fail or 
  its quality can deteriorate, thus 
  making impossible to a particular QoSN to maintain the negotiated quality. In 
  that case, the interested QoSN can enhance the throughput of its link via stream 
  striping on non-corouted paths. In this 
  case, it sends back a message to temporarily stop the stream, and sends forward 
  a message to suspend updates in AC tables on the path. Then, it sends its clones 
  to handle new non-corouted paths and starts the negotiation phase with the clones. 
  When negotiation completes, the QoSN sends back a message that restarts the 
  stream: apart from a delay in receiving the stream, the VoD target goes on transparently. 
  
 In the case of multicast distribution 
  of the same video stream (for N targets), the generated network traffic can 
  be limited by exploiting location awareness of agents. While in 
  traditional VoD systems the source generates N packet streams, one for each 
  target, with the consequent overload, our approach can take advantage of locality. 
  Our QoSNs can ascertain whether there are several targets within the same domain 
  locality, and split packets only when it is necessary, in general only at the 
  last hop. This is commanded by the QoSN at the gateway of the last domain.
 
  In the case of multicast distribution 
  of the same video stream (for N targets), the generated network traffic can 
  be limited by exploiting location awareness of agents. While in 
  traditional VoD systems the source generates N packet streams, one for each 
  target, with the consequent overload, our approach can take advantage of locality. 
  Our QoSNs can ascertain whether there are several targets within the same domain 
  locality, and split packets only when it is necessary, in general only at the 
  last hop. This is commanded by the QoSN at the gateway of the last domain. 
See the next page for first performance evaluations of the SOMA VoD service.
|  | |||
| Page 
          updated on  | In 
          case of problems, or if you find any bug, please contact us. | ||