dtrace discuss - Mar 2006 - Pathetically new to dtrace but looking for help with profiling a system

I''m pathetically new to dtrace :) and yes I plan to look over the
dtrace tutorials and such.

I''m curious if anyone has already done something like this with dtrace.
I''m looking to profile a system as it is getting hammered with network
traffic and doing disk updates/reads, etc.  What I would like to do is be able
to say for a population of #### concurrent users all doing somewhat similar work
application XYZ is consuming #% of the system, its disk IO is consuming another
#% of the system, and its network traffic is consuming #%.  The network traffic
is rather outbound heavy with the inbound traffic being about 56% of the
outbound.  So if outbound is 1Mb/sec the inbound will be 560Kb/sec per session
of application XYZ.  1Mb/sec is only an example, the actually numbers could be
MUCH higher.  So profiling CPU consumption by the network is very important to
me.

The network traffic has a flip side on the receiving server.  I want to be able
to profile the receiving server as well.

Some of the sample dtrace scripts I''ve seen on opensolaris.org look
very interesting and could probably help with this effort but I don''t
believe they answer the above.

So am I over complicating this or is this doable ?
This message posted from opensolaris.org

On Sat, 18 Mar 2006, Sean O''Neill wrote:
> I''m pathetically new to dtrace :) and yes I plan to look over the
dtrace tutorials and such.
>
> I''m curious if anyone has already done something like this with
dtrace.  I''m looking to profile a system as it is getting hammered with
network traffic and doing disk updates/reads, etc.  What I would like to do is
be able to say for a population of #### concurrent users all doing somewhat
similar work application XYZ is consuming #% of the system, its disk IO is
consuming another #% of the system, and its network traffic is consuming #%. 
The network traffic is rather outbound heavy with the inbound traffic being
about 56% of the outbound.  So if outbound is 1Mb/sec the inbound will be
560Kb/sec per session of application XYZ.  1Mb/sec is only an example, the
actually numbers could be MUCH higher.  So profiling CPU consumption by the
network is very important to me.
>
> The network traffic has a flip side on the receiving server.  I want to be
able to profile the receiving server as well.
>
> Some of the sample dtrace scripts I''ve seen on opensolaris.org
look very interesting and could probably help with this effort but I
don''t believe they answer the above.
>
> So am I over complicating this or is this doable ?
> This message posted from opensolaris.org
Hi Sean,

I don''t think you will find a single script to provide this
information,
but Brendan''s DTraceToolkit provides a good deal of information on the 
areas you are interested in:

http://users.tpg.com.au/adsln4yb/dtrace.html#DTraceToolkit

The DTraceToolkit hash scripts to profile applications, and the disk and 
network scripts are top notch (answering network stack specific questions 
can be done with the fbt and mib providers). You could also try posting 
to the opensolaris-network discussion group, since they were supposedly 
working on a DTrace network provider.

Hope this helps,
- Ryan
--
UNIX Administrator
http://daemons.net/~matty

You talk   about #% of  a  system, is this   measured in cpu
utilization or  something else ? Your best  bet to track cpu
utilization is to use the profile provider and to do a study
as you describe you may want to try er_kernel layered on top
of that profile provider.

	http://developers.sun.com/prodtech/cc/articles/perftools.html

-r




Sean O''Neill writes:
 > I''m pathetically new to dtrace :) and yes I plan to look over the
dtrace tutorials and such.
 > 
 > I''m curious if anyone has already done something like this with
 > dtrace.  I''m looking to profile a system as it is getting
hammered
 > with network traffic and doing disk updates/reads, etc.  What I would
 > like to do is be able to say for a population of #### concurrent users
 > all doing somewhat similar work application XYZ is consuming #% of the
 > system, its disk IO is consuming another #% of the system, and its
 > network traffic is consuming #%.  The network traffic is rather
 > outbound heavy with the inbound traffic being about 56% of the
 > outbound.  So if outbound is 1Mb/sec the inbound will be 560Kb/sec per
 > session of application XYZ.  1Mb/sec is only an example, the actually
 > numbers could be MUCH higher.  So profiling CPU consumption by the
 > network is very important to me. 
 > 
 > The network traffic has a flip side on the receiving server.  I want
 > to be able to profile the receiving server as well. 
 > 
 > Some of the sample dtrace scripts I''ve seen on opensolaris.org
look
 > very interesting and could probably help with this effort but I
don''t
 > believe they answer the above. 
 > 
 > So am I over complicating this or is this doable ?
 > This message posted from opensolaris.org
 > _______________________________________________
 > dtrace-discuss mailing list
 > dtrace-discuss at opensolaris.org