dtrace discuss - Apr 2008 - DTrace usage in the PostgreSQL database

New blog entry using DTrace in PostgreSQL to find scalability issues. Focus is
more on DTrace as an instrumentation tool.

Please check
http://blogs.sun.com/paulvandenbogaard/category/PostgreSQL

Happy reading, and open for feedback. So ideas or thoughts? Do let me know.


--
This message posted from opensolaris.org

Hi, Paul,

This is really informative blog! Is there any arguments in your USDT probes you
measured and how''s their perf impacting vs. no arguments? and
how''s perf impact of using USDT comparing to use pid provider to get
the similar info?

Is there future plan working on reduce the instrusiveness of dtrace? I think
this is critical for DTrace''s utilization in the production
environment.


Thanks,

   Luojia(Jenny) Chen              
   Software Engineer                 
   ISV Engineering      
   v-mail:(510) 550-2394                                  
   SUN Microsystems

----- Original Message -----
From: Paul van den Bogaard <Paul.Vandenbogaard at Sun.COM>
Date: Thursday, April 3, 2008 9:29 am
Subject: [dtrace-discuss] DTrace usage in the PostgreSQL database
To: dtrace-discuss at opensolaris.org

> New blog entry using DTrace in PostgreSQL to find scalability issues. 
> Focus is more on DTrace as an instrumentation tool.
>  
>  Please check
>  http://blogs.sun.com/paulvandenbogaard/category/PostgreSQL
>  
>  Happy reading, and open for feedback. So ideas or thoughts? Do let me 
> know.
>  
>  
>  --
>  This message posted from opensolaris.org
>  _______________________________________________
>  dtrace-discuss mailing list
>  dtrace-discuss at opensolaris.org
>

On Thu, Apr 03, 2008 at 10:48:41AM -0800, Luojia Chen
wrote:
> Is there future plan working on reduce the instrusiveness of dtrace?
> I think this is critical for DTrace''s utilization in the
production
> environment.
We''ve taken steps to reduce the disabled probe-effect in particular
with
is-enabled probes. If there''s some benchmark you''re using
which indicates
that the presence of USDT probes is hurting performance, we''d love to
see
it.

Adam

-- 
Adam Leventhal, Fishworks                        http://blogs.sun.com/ahl

Jenny,

these tests did not focus on the "number of arguments" effect. And
indeed most of these probes have two parameters (lockid and lock mode), a couple
have four.

We are currently building a small statistics keeping module that stores process
stats of any kind in shared memory so an external program can monitor it. Will
replace the USDT in use by call to functions from this module.  Next step is to
compare the DTrace results with these results.
As I said in my blog, I do not care if there is a strong instrumentation effect
as long as the behaviour of the application stays the same. I just want to make
sure this is the case so our conclusions will be valid.


--
This message posted from opensolaris.org

just added another entry to this blogs PostgreSQL category. More findings with
DTrace on PostgreSQL but also on DTrace.

Happy reading. Ideas, thoughts?  Please do let me know!


--
This message posted from opensolaris.org