| Title | : | Proxy Server based Control Mechanisms for Abusive Internet Traffic |
| Speaker | : | Saad Yunus Sait (IITM) |
| Details | : | Tue, 29 Sep, 2015 2:00 PM @ BSB 361 |
| Abstract: | : | Peak-hour congestion is a common problem faced by institutions accessing the In-
ternet. ISPs have traditionally used schemes like usage-based pricing (UBP) and time-
dependent pricing (TDP) to control subscriber traffic, but control within an organization
is outside the purview of an Internet Service Provider. Most methods of control within
an organization are node-based, so they fail when nodes are shared by users and when
IP addresses change due to dynamic host configuration protocol(DHCP), as in the case
of mobile devices. Besides, control is normally done by identifying ’rogue’ nodes which
generate too much traffic and placing them in a low priority queue when their quota is
exceeded. This threshold-based scheme is inaccurate because it places a hard threshold
on the usage, whereas Internet traffic is dynamic and access patterns vary from day to
day. We use a combination of two approaches to control abusive Internet access. While machine learning techniques are used to identify abusive usage and categorize users, TCP-based control mechanisms are used to control users. We have assumed the presence of a user-authenticating forward web proxy through which users connect to the Internet; as a result, our control is user-based as opposed to node-based. We illustrate our scheme to control abusive users by manipulating the advertised window, pacing out the ACKs, and using admission control mechanisms. We show that our scheme is able to control flows belonging to abusive users, while improving the experience of normal users; response times of normal users are better by 20% and throughputs of normal large flows are better by 200%; the overall performance of the system is also preserved. Our scheme also provides the benefit of reducing bandwidth wastage due to user impatience. The scope of our work is limited to the control of download traffic. |
