LOB Operation and SQL*Net Message From Client and cursor #0

  • From: "Larry Elkins" <elkinsl@xxxxxxxxxxx>
  • To: "'Oracle-L'" <oracle-l@xxxxxxxxxxxxx>
  • Date: Mon, 29 Apr 2013 20:22:44 -0500 EE on AIX 6.1

Any known instrumentation issues / oddities when fetching LOBS?

Tracing a process, 10046 level 8, 91% of the time on sql*net message from 
client, with 99.8% of that accumulated under cursor #0.
Retraced with event 10051 turned on, indicated they were OPI call type 96, 
Lob/FILE operations. 

So, simple enough, a little over a million calls back on forth on those OPI 
level calls to get the LOB data. Precious little DB

But, techs with opnet installed and gathering data in different tiers swear 
there is little time on the network between the app
server and the unix host where the database resides, that the time is *in* the 
DB. He's measuring nic to nic and acknowledgments
between the app server and db server host, at least that's the way he described 
it. DB host is a VM. 

So first step is to make sure there isn't some sort of instrumentation bug in 
Oracle related to LOBS, and that the time I'm showing
on sql*net message from client is indeed just that, and not something like 
"uncounted" CPU. And if it is indeed sql*net message from
client, then we can start to look at where that time he shows is "in the 
database" actually is on that host.

Larry G. Elkins


Other related posts: