6. An ISP streaming traffic management algorithm
In Section 4 we introduced a simple game theoretic framework but in all the models we presented the equilibrium points
can only be achieved via common knowledge. In other words, each ISP needs to know the streaming server upload capacity
and the Θi-s of all the other ISPs. Obviously this hypothesis is not realistic in practical systems. Nevertheless, by using the
ideas presented in Section 5 we can define a distributed algorithm that implements the EGT version of the game without
the constraints that the ISP must be aware of the Θi-s of all the other ISPs.
The proposed distributed algorithm could be implemented by means of a two tier structure of trackers (similar in the
spirit to the tracker architecture proposed in [9,22,17]). At the top level there is the super-tracker and at the lower level
there is a collection of ISP-trackers (one for each ISP). Each ISP-tracker must be able to measure the current values of (i) the
number of active peers in the ISP, (ii) the number of peers that receive the video at rate r, and (iii) the values for Θi
, and Φi
.
Furthermore, the ISP-tracker is able to modify the bandwidth offered to the other ISPs (e.g., by using bandwidth throttling).
Each ISP-tracker periodically (e.g., each δ sec) sends the collected measures to the super-tracker that can use them to
verify whether the system is in a universal streaming region or not. The super-tracker replies to the ISP-trackers with a
universal-streaming or no-universal-streaming message. In turn, the ISP-trackers are required to update the
values of Φi-s according to Eq. (5).
This two-tier architecture represents a practical solution for using global knowledge at the ISP level. Obviously, the ISPs
shall trust the common super-tracker, that is actually acquiring ISP-trackers private information but it is not disclosing
such data to the ISPs. In fact the super-tracker only discloses the result of the universal streaming indicator function. In
this paper we neglect a number of open issues that are left for future research and practical implementation. Among such
extensions we mention the need for incentive mechanisms to enforce cooperation among the ISPs, the implementation
control techniques to guarantee ISP honesty, and the confidentiality of the communications between the super-tracker and
ISP-trackers. Finally it is important to notice that the two-tier architecture, composed of the super-tracker and the collection
of ISP-trackers, only aims at ISP traffic management and does not replace or interfere with the architecture of the adopted
P2P application. In other words, if the P2P streaming architecture uses a tracker similar to the BitTorrent protocol to assist
the streaming distribution (e.g., to provide peer addresses) this tracker does not interfere with the ISP traffic management.
6.1. Experimental results
To investigate the characteristics of the EGT scheme we develop a discrete event simulator that implements in an abstract
manner most of the features of a pull-based P2P streaming application that interacts with our EGT algorithm.
Simulator description. The simulator models a P2P streaming system where the peers are grouped into k different ISPs. The
simulator is based on the implementation of three agents, namely the peer, the ISP-tracker and the super-tracker. The