Submitted by Matthew Khouzam on Thu, 04/19/2012 - 14:13
I was just thinking the following items monthly reporting, this is an aspect of my job that I do enjoy very much, just not as much as all the other aspects my job. :)
These things can help me and anyone else doing the reports make them easier to write.
Things I like seeing in monthly updates:
* Milestones, no matter how trivial, like: Bob reviewed the following papers. Alice designed a new ringbuffer that works on single core computers.
Submitted by Francis Giraldeau on Wed, 03/28/2012 - 10:05
Paradyn Week 2012 was held at the University of Maryland. I went there to see what other folks were doing in the area of debugging and runtime software monitoring with executable binaries modification. I wanted to compare that approach with our current tracing methods.
Submitted by Raphael Beamonte on Thu, 03/22/2012 - 01:55
As I was working on "patching" the hwlat_detector kernel module to allow using it on kernel 3.0+ and didn't found any place to publish my work, I decided to let it available on a git repository.
You'll be able to find it here : https://github.com/XaF/hwlat-detector
Submitted by Francis Giraldeau on Wed, 03/21/2012 - 10:19
With the newest release of LTTng 2.0, we did an experiment about buffer size for kernel tracing. This experiment is necessary to verify if there exists a closed loop tracing. The consumer daemon of LTTng writes the trace on disk, then new events are being generated. If enough events are generated while flushing a buffer and fills another buffer, then the system may enter closed loop tracing.
Submitted by Julien Desfossez on Tue, 03/20/2012 - 12:28
For the past weeks along with the preparation of the stable release of LTTng 2.0 (especially the Babeltrace API part) and LTTngTop development, I have been experimenting with Google Rocksteady.
This application is the monitoring solution used at Google, so it is designed to be highly scalable and distributed.
Submitted by Yannick Brosseau on Sun, 03/04/2012 - 14:04
Submitted by Francis Giraldeau on Wed, 02/29/2012 - 09:24
Not much this week about new research stuff, only code refactoring. The problem was that network objects were clunky and it was pretty hard to retreive them based on various criteria. It ended in many lookup loops that was not scaling well. For each search type, a HashMap is now used and reduces the running time. There is a generic object used to hold the history of replaced objects. This will be interesting how to map this data structure to state history backend in the future.
Submitted by Julien Desfossez on Tue, 02/28/2012 - 22:04
A long time passed since my last update...
I have now started my Ph. D. I am still focusing on using the tracing as a useful and efficient problem-solving facility, but this time it will deal with the scalability. One of the early goal being to use the tracing in large data center as a highly granular monitoring interface. A lot remains to be defined, but that is the general idea.
Submitted by Raphael Beamonte on Tue, 02/28/2012 - 18:55
Submitted by Francis Giraldeau on Wed, 02/22/2012 - 09:55
Nfqueue has been used to simulate network delay. Traffic shaper adds delay only before transmitting a packet, while a script intercepting packets with nfqueue can add delays both to reception and transmission.
The test case has been traced with nfqueue adding delay of 50ms. The results has been compared with TC delay of 100ms. The same blockings has been observed, but an order of magnitude of increased delay has been observed for nfqueue. More analysis must be done to highlight the source of the difference.