Re: LTTng and network output

Hi Andrej,

* Andrej van der Zee ([email protected]) wrote:
> Dear Mathieu Desnoyers,
> 
> With great interest I have studied the LTTng project
> for the last couple of days. 

I'm glad of the interest you show for my project :)

> 
> I am a PhD candidate at Waseda University, Tokyo, and
> interested in using LTTng for our own research
> purposes. For our research, we would like to do
> real-time analysis of events generated by a cluster.
> Well, in a nutshell. Therefor we have to send the
> events to a central server in real-time. I found a
> discussion between you and Jens Axboe from blktrace
> from last year where you talked about a way to send
> the events over the network without copying it between
> users/kernel-space (twice), by means of sendfile() or
> mmap()+send(). Also, in one of your papers you mention
> that lttd and libltt-usertrace-fast "should eventually
> be extended to other media like network
> communication".
> 
> I was wondering if there is already work done on a
> lttd-like daemon that writes to the network instead of
> disk files (forgive me if this question already popped
> up in the mailing lists).
> 

Not implemented yet. The idea would be the following :

- "Client" version of lttd which sends files over the network.
  Basically, the send call would read from the mmapped buffers and 
  write to a socket file descriptor that would be connected to a
  lttd-server. Note that there would be a more efficient way to do this,
  which involves using the sendfile() file operation, which is already
  implemented in relay.c (the latter is the preferred way to do it.).
- If you want to have a stream of information somehow "guaranteed" to be
  sent at a certain interval, you will probably want to create a new
  timer in the kernel that calls the "subbuffer switch" with "ACTIVE"
  mode : it will switch subbuffers, which will make the current
  subbuffers ready for delivery. If you do not do this, the low-rate and
  medium-rate buffers may not be delivered to the server until the end of
  tracing. Ideally, the interval of the timer and the activation of this
  feature should be controlled by ltt-control.ko/lttctl.

Contributions would be welcome :)

Best regards,

Mathieu




> My best regards,
> 
> Andrej van der Zee
> 
> 
> 	
> 	
> 		
> ___________________________________________________________ 
> New Yahoo! Mail is the ultimate force in competitive emailing. Find out more at the Yahoo! Mail Championships. Plus: play games and win prizes. 
> http://uk.rd.yahoo.com/evt=44106/*http://mail.yahoo.net/uk 

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68