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

  • From: "Jonathan Lewis" <jonathan@xxxxxxxxxxxxxxxxxx>
  • To: "'Oracle-L'" <oracle-l@xxxxxxxxxxxxx>
  • Date: Thu, 9 May 2013 13:35:34 +0100

| Had to put this in the backburner for a bit, but had already tired 400 
for long and longc with no impact, and tried with 4000 after
| your suggestion. The round trips remained at 534,073 for all the tests.

I must have lost track of the numbers - I thought you had said something 
about 100,000 rows and 359,000 round trips.
I was expecting a change in longchunksize and long to reduce this to at 
most 2 round trips per row.


Jonathan Lewis

Author: Oracle Core (Apress 2011)

----- Original Message ----- 
From: "Larry Elkins" <elkinsl@xxxxxxxxxxx>
To: <jonathan@xxxxxxxxxxxxxxxxxx>; "'Oracle-L'" <oracle-l@xxxxxxxxxxxxx>
Sent: Saturday, May 04, 2013 2:30 PM
Subject: RE: LOB Operation and SQL*Net Message From Client and cursor #0

| Copied the table to an DB sitting on a non-VM database server, 
SQL*Plus with the 4000 settings. I've made sure the table
| was cached for all these tests to eliminate PIO differences (which I 
would be able to account for anyway since the would be recorded
| in the trace). So, for both 10g VM and 11g non-vm, here's what we would 
| TRIPS:          534,073
| ROWS:           300,700
| LOBREADS:       233,700 (66,700 null nclobs)
| LOBREAD TIME:   5.918873 seconds, median 24 microseconds, avg 25.3268 
microsecond, with a few outlier on the high side, and a
| fastest of 23 microseconds.


Other related posts: