|M.Sc Student||Getselevich Vladimir|
|Subject||Enabling Efficient Fast Track Services in Active Networks|
|Department||Department of Computer Science||Supervisor||Professor Dan Raz|
Active networks are a new approach to network architecture. Routers can perform computations on user data, while packets can carry programs to be executed on routers and possibly change their state. However, while the ability to perform additional computations in intermediate nodes indeed adds powerful functionality to the network, at the same time it could add a significant delay to the basic packet forwarding operation performed at the routers. Basic router forwarding functions are fast because they analyze only packet headers and generally implemented using fast hardware devices, whereas active network modules are usually implemented using slower software modules. In addition, the forwarding of an entire packet from the router to the active network module through hardware/software or kernel/user level boundaries also takes time.
Thus, an important challenge in realizing active networking inside network forwarding nodes is to accommodate the specific needs of some flows without penalizing all flows and to enhance the efficiency of packet flow processing.
In this research we study several operators that can efficiently be applied to the data flows in order to reduce the delay of the router critical forwarding path. The studied operators (also called primitives) are: Tee - making copy of data packets and forwarding them to the active modules without interfering with the fast delivery of the original data packets;
Partial Frame Peeking - copying only a small fraction of the data packet to be forwarded to the active module, and Statistical Sampling - making a selective probabilistic processing of packets, in a way that only a certain part of the selected flow packets will be forwarded to the active module. We extended the functionality of ABLE, an active network system, by adding implementation of these new operators.
In order to demonstrate the advantage of the new operators we developed several active applications and tested their performance in the LCCN active network test bed. We also proposed an improvement for several active network algorithms and performed theoretical analysis of their time and message complexity using a recently developed formal model.