Reading for Lecture Nov 13

X-trace: A Pervasive Network Tracing Framework

When complex systems misbehave, it is quite difficult to diagnose the source of the problem. X-trace is a framework that provides a comprehensive view of systems that adopt it. A user or operator invokes X-trace when initiating an application task, by inserting X-trace metadata with a task identifier. X-trace is comprehensive in the sense that it tags all network operations resulting from a particular task with the same task identifier, this set is called a task tree.

Diagnosing problems often requires tracing a task across different administrative domains (AD). Since ADs may not want to reveal information to each other, X-trace provide a clean separation between the client that invokes X-trace and the recipient of the trace information.
The design principles of X-trace are: 1) The trace request should be sent in-band, rather than in a separate probe message. 2) The collected trace data should be sent out-of-band, decoupled from the original datapath. 3) The entity that requests tracing is decoupled from the entity that received the trace report.

The task identifier within X-trace metadata is uniquely identifies each network task. The pushDown() and pushNext() operations are used to propagates X-trace metadata along with the datapath. The reports at each X-trace aware node are used to reconstruct the datapath.

End-to-End Internet Packet Dynamic

Each 100 Kbyte transfer at both the sender and receiver to distinguish between the end-to-end behaviors due to different directions on Internet paths, which often exhibit the asymmetries. Various issue is investigated in the paper such that out-of-order packet delivery, packet corruption, bottleneck bandwidth, pattern of packet loss, …

Two runs are used to find out the dynamic packet change in the course of 1995. Lots of different this are observed, discussed and inferred but it is not clear for me what are the main points.