This paper is mainly targeted for real-time applications. The authors propose that real-time traffic require guaranteed service or predicted service (adjust themselves based on the network delay). The architecture discussed handles both types of traffic.
Real-time applications are sensitive to data delivery delay, require delay information to set playback point, data should be delivered before its playback point and can tolerate only a certain fraction of packet losses. It is required that real-time applications get service as if they are the only one using the link and they should not be affected by other application traffic. Service commitments made to an end-user depends on the traffic commitment of the end-user.
For guaranteed traffic it is proposed to use token bucket filters to handle the rate of traffic while WFQ is used for scheduling. For the predicted services the authors propose FIFO+. To make both these algorithms work together guaranteed service flows are assigned as different WFQ flows while datagram/predicted service flows are assigned a pseudo flow 0. This flow has different priority classes and each class runs FIFO+.
The paper puts forward the need and requirement for servicing real-time application traffic. The discussion of predicted service shows that not all of the real-time traffic requires guaranteed service. One of the main weaknesses of the paper is the lack of concrete evaluation. Also, it has not been discussed how deployment will be handled. Techniques that propose changes to the Internet should always keep in mind the ease of deployment and the incentive for people to adopt the scheme.
Real-time applications are sensitive to data delivery delay, require delay information to set playback point, data should be delivered before its playback point and can tolerate only a certain fraction of packet losses. It is required that real-time applications get service as if they are the only one using the link and they should not be affected by other application traffic. Service commitments made to an end-user depends on the traffic commitment of the end-user.
For guaranteed traffic it is proposed to use token bucket filters to handle the rate of traffic while WFQ is used for scheduling. For the predicted services the authors propose FIFO+. To make both these algorithms work together guaranteed service flows are assigned as different WFQ flows while datagram/predicted service flows are assigned a pseudo flow 0. This flow has different priority classes and each class runs FIFO+.
The paper puts forward the need and requirement for servicing real-time application traffic. The discussion of predicted service shows that not all of the real-time traffic requires guaranteed service. One of the main weaknesses of the paper is the lack of concrete evaluation. Also, it has not been discussed how deployment will be handled. Techniques that propose changes to the Internet should always keep in mind the ease of deployment and the incentive for people to adopt the scheme.
No comments:
Post a Comment