POPI Project

[intro] [methodology] [software] [feedback] [people]


Packet forwarding prioritization (PFP) has been available in off-the-shelf routers like Cisco and Juniper along with some kinds of middleboxes such as traffic shapers. ISPs have come to rely on these mechanisms for managing their networks, for example as a way of rate limiting certain classes of applications (e.g. peer-to-peer), see an article in WSJ.

PFP may have a significant impact on the performance of your applications. It can also impact the accuracy of measurement tools' output and the effectiveness of network troubleshooting procedures. For example, when there is ICMP rate-limiting, ping may not accurately represent the loss rates of TCP applications. However, users, developers and most other network administrators have no information of such settings nor ways to procure it. Therefore, we provide POPI (Packet fOrwarding Priority Inference) as an end-to-end tool to infer such settings.

Our methodology is simple. We measure the path with different types of packets and then compares the loss rates of them. We designed a robust statistical method, rank method, for comparing the loss rates of different packet types.

methodology figure

To use POPI, you should specify at least two different packets, choose a probe mode, and specify the destination IP. POPI currently works in two probe modes, end-to-end mode and hop-by-hop mode. (see README in the software package). It currently allows you to specify: Download: here
This research was conducted jointly by Tsinghua NGN and Northwestern LIST. Feedback, comments and bugfixes are welcomed. Contact Guohan Lu lguohan email.
This work is partially supported by China 863 Program under grant 2006AA01Z201130 and by USA Department of Engergy (DoE) Early Career Award DE-FG02-05ER25692.

Valid HTML 4.01 Transitional