|
Description
|
From an email conversation with Adam Leventhal:
----------
>I was playing around with the example posted on
>> http://blogs.sun.com/roller/page/fintanr/20040909#dtrace_and_how_much_data,
>> and it isn't actually correct (I've mailed him to let him know). If you
>> use his mib probe alongside one that sums the bytes written by the
>> process the two don't agree. From further poking around it becomes
>> obvious that the mib probes don't necessarily fire at the same time as
>> the IO event that generates the increment. It appears that they mostly
>> fire during the scheduler tick, and most of the time the target process
>> (in his case httpd) isn't the one on proc at this point. It might be
>> worth pointing this out in the docs...
Good point. I'm cc'ing the DTrace core alias and we'll try to get this
resolved as we work on the docs. Feel free to file a bug in doc/dtrace
if you'd like to track this.
----------
The example D script in question is:
dtrace -n mib:::tcpOutDataBytes'/execname == "httpd"/{@[execname] = sum(args[0])}'
xxxxx@xxxxx.com 2004-09-25
This seems to be a common misconception - see also http://blogs.sun.com/roller/comments/akolb/Weblog/solaris_10_and_mib_counters
|