DTrace Meeting Notes 03 March 2016
Attending: Shrikanth Kamath, Robert Watson, Graeme Jenkinson, Arun Thomas, Mark Johnston, Erin McNeil, George Neville-Neil
Introductions all around.
Mark has no plans to add to DTrace of late. He has worked on the SDT probes to do zero cost probes. He can put that into a Phabricator. Robert asks if he'll need help getting this to work on other architectures. Mark will need help on ARM but Robert can point people at Cambridge at that.
Erin is on a performance team that is trying to measure their system. They've been playing with DTrace under large scale situations they've seen issues. Shrikanth points out that they have a module for the network stack and that aggregations are having problems with performance.
Robert mentions the NetSem work on TCP. It would be good if other folks were interested in using the tcpdebug output. Shrikanth with check with the folks at Juniper that work on the network stack.
Robert describes some of the work we're doing with CADETS, including machine readable output, packet capture integration and distributed work. Graeme asks for good use cases for distributed systems.
Mark thinks that converting to DWARF from CTF would be a big job. One of the Illumos folks has a command that will do this on the fly. Mark would like DTrace to have more information about the source code, but that may require DWARF conversion.
Robert says that it would be best if everything could be traced when shipped.
George mentions the idea of OpenDTrace. Mark points out that much of the issues with DTrace have to do with the fact that it knows the "entrails" of the kernel. Robert explains how openbsm works, which is the model that OpenDTrace wishes to take. Robert points out that the DIF engine can be used outside the kernel, in particular in a program somewhere.