From owner-freebsd-questions  Fri Oct  3 00:00:06 1997
Return-Path: <owner-freebsd-questions>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id AAA25025
          for questions-outgoing; Fri, 3 Oct 1997 00:00:06 -0700 (PDT)
Received: from primer.i-connect.net (primer.i-Connect.Net [206.190.143.13])
          by hub.freebsd.org (8.8.7/8.8.7) with SMTP id XAA24977
          for <questions@freebsd.org>; Thu, 2 Oct 1997 23:59:59 -0700 (PDT)
Received: (qmail 897 invoked by uid 1008); 3 Oct 1997 07:04:34 -0000
Mailing-List: contact freebsd-dlm-help@primer.i-connect.net; run by ezmlm
Delivered-To: mailing list freebsd-dlm@primer.i-connect.net
Received: (qmail 891 invoked from network); 3 Oct 1997 07:04:33 -0000
Received: from sendero-ppp.i-connect.net (206.190.143.100)
  by primer.i-connect.net with SMTP; 3 Oct 1997 07:04:33 -0000
Received: (qmail 18461 invoked by uid 1000); 3 Oct 1997 07:00:22 -0000
Message-ID: <XFMail.971003000022.Shimon@i-Connect.Net>
X-Mailer: XFMail 1.2-alpha-092597 [p0] on FreeBSD
X-Priority: 3 (Normal)
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
MIME-Version: 1.0
Date: Fri, 03 Oct 1997 00:00:22 -0700 (PDT)
Organization: Atlas Telecom
From: Simon Shapiro <Shimon@i-Connect.Net>
To: freebsd-hackers@freebsd.org, freebsd-dlm@primer.i-connect.net
Subject: Advice Seeked - Non-Invasive Metrics Techniques
Sender: owner-freebsd-questions@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Hi Y'all,

I hope the subject line is accurate.

I need to carefully tune and analyze timing and perfromance statistics.
Mainly to identify why some rotines in the DLM take so much time.

I am looking for ideas on how to measure the behavior of code, take time
measurements, count events, etc. without consuming so much time.

For example, when measuring time, I have the following strategy:

In the dlm_lock_t structure, I have

typedef struct dlm_lock
}
...

    struct timeval start_time, stop_time;

} dlm_lock_t

Then, I have the function:

static INLINE void
dlm_debug_times(dlm_lock_t *lock, int line)
{
    microtime(&lock->stop_time);
    printf("So far (%d) %d\n", line,
           time_delta(lock->start_time, lock->stop_time));
}

...

Then, when I want to time sone code I do

static int
some_function(dlm_lock_t *lock, ...)
{
    ...

    microtime(&lock->start_time);

    ...

    /* Before a certain block of code */
    dlm_debug_times(lock, __LINE__);

    ...
    /* Some code to be timed */
    ...

    dlm_debug_times(lock, __LINE__);

    ...
}

Is there a better way to do that?
In this context, better means a way to impact the performance of the system
to a lesser degree.

I have a sneaking suspicion that the above mechnism takes a lot of time,
thus upsetting the code being timed.

Any suggestion will be appreciated.

Thanx!


---


Sincerely Yours, 

Simon Shapiro                                                 Atlas Telecom
Senior Architect         14355 SW Allen Blvd., Suite 130 Beaverton OR 97005
Shimon@i-Connect.Net                                  Voice:   503.799.2313