RE: N/w Bottleneck Check under Benchmark Load & Event "SQL*Net more data from client"

From: Mark W. Farnham <mwf_at_rsiz.com>
Date: Tue, 1 Jul 2008 12:51:22 -0400
Message-ID: <053d01c8db9a$b4080490$1100a8c0@rsiz.com>


That's almost certainly a useful thing to do, but we cannot make a prediction how much it will increase the throughput accomplished in a comparable 9 minute window, because those statistics are aggregates of all the sessions running and many of the waits are almost certainly overlapped with respect to real time. But since the effort required is minimal and it is likely to reduce the noise in your measurements and marginally reduce the elapsed time of each individual transaction, it is probably a good move.  

Now we don't know if any of those individual waits were long, but even if each was exactly the average, you're talking about less than half a millisecond line turnaround plus data transmission.  

Since it is a "more" data request (which keyword I had missed until Cary's response), it wouldn't be user think time as any of the delay - but it could be client side program context switching.  

It would be interesting to know what your bottleneck(s) are that are proportional to real elapsed time, and if you do optimize packet size and these waits dramatically disminish, if there is an increase in throughput over a comparable amount of real elapsed time. There is a non-zero chance that reducing these waits will decrease your throughput if it exacerbates the hypothetical server bottleneck that truly gates throughput on the database server. (Though that seems unlikely in this case.)  

Nice exact pointer, Yechiel!


From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Yechiel Adar
Sent: Tuesday, July 01, 2008 12:09 PM
Cc: 'ORACLE-L'
Subject: Re: N/w Bottleneck Check under Benchmark Load & Event "SQL*Net more data from client"  

It seems to me that you need to send larger packets to the server. Increase the SDU parameter and try again.

Check out metalink note:44694.1
SQL*Net Packet Sizes (SDU & TDU Parameters).

Adar Yechiel
Rechovot, Israel

VIVEK_SHARMA wrote:

Folks  

For a OLTP Benchmark under Transaction Load is event "SQL*Net more data from client" an indication of Network Choke

Currently a 100 MBPS LAN Cable connects APP to DB Server.  

Oracle 10.2.3.0 (NON-RAC)

Application - Banking Java Application using Oracle thin driver to connect to the Database.

HP-UX 11.11  

Cheers & Thanks V much  

Vivek

P.S.  

                 Snap Time

Begin Snap: 30-Jun-08 19:17:41

End Snap: 30-Jun-08 19:26:43

Elapsed: 9.03 (mins)

DB Time: 9.06 (mins)    

Top 5 Timed Events                                                    Avg
%Total
~~~~~~~~~~~~~~~~~~                                                   wait
Call
Event                                            Waits    Time (s)   (ms)
Time Wait Class
  • ------------ ----------- ------ ------ -------
SQL*Net more data from client                            338,920
149                           0      20.3    Network




--
http://www.freelists.org/webpage/oracle-l
Received on Tue Jul 01 2008 - 11:51:22 CDT

Original text of this message