dtrace discuss - Aug 2006 - CPU utilization with ring buffer

I have a simple script, adapted from something I found online, to report file
read and writes, along with FD and a timestamp. dtrace runs the script and uses
very little CPU ~2%. If I change the bufpolicy to ring and run the script dtrace
CPU usage jumps to near 50% on a 2 CPU system (as reported by prstat). Is there
something in my script that causes this problem, and hints? - Chris
-------------

#!/usr/sbin/dtrace -s 
#pragma D option bufpolicy=switch
#pragma D option bufsize=10M  
#pragma D option bufresize=manual
#pragma D option quiet

syscall::read:entry
/execname != "dtrace"/
{
    self->event = "READ";
}

syscall::write:entry
/execname != "dtrace"/
{
    self->event = "WRITE";
}


syscall::read:entry, syscall::write:entry
/execname != "dtrace"/
{
   
printf("%s\",\"FS\",\"%s\",\"%d\",\"-\",\"%d\",\"\n",
execname, self->event, arg0, timestamp);
    self->event = 0;
}
 
 
This message posted from opensolaris.org

Looks like this:

	http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6253027
	http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=2129582

That was fixed in Nevada and S10U1. Which OS are you running ?

Chris LaRosa writes:
 > I have a simple script, adapted from something I found online, to
 > report file read and writes, along with FD and a timestamp. dtrace
 > runs the script and uses very little CPU ~2%. If I change the
 > bufpolicy to ring and run the script dtrace CPU usage jumps to near
 > 50% on a 2 CPU system (as reported by prstat). Is there something in
 > my script that causes this problem, and hints? - Chris