Re: SQL*Net more data from client description

On Jan 27, 10:02 am, "Jonathan Lewis" <jonat..._at_jlcomp.demon.co.uk> wrote:
> Steve,
>
> Coming in late, here, but the doc does seem to
> have a 'cut and paste' bug.
>
> More data from client isn't about MTU (which is
> the transport unit of the network) but about the
> Oracle's SDU (session data unit).
>
> Oracle chats in sizes of the SDU (ca. 2000 bytes
> I think), and waits for acknowledgement from the
> far end. The network may chat in smaller units,
> breaking the Oracle message into a couple of chunks
> (typical MTU is about 1470 bytes) - but the network
> does not wait for acknowledgement between the two
> packets.
>
> "More data" means that there is more data to send than
> the SDU can hold. An insert of a varchar2(4000) could
> be sufficient for this to appear.
>
> You can increase the SDU size by fiddling in the listener.ora,
> tnsnames.ora and sqlnet.ora - the specific details of method
> varies with version. In the latest version, the easiest option
> is to put
> default_sdu_size = NNNNN
> in the sqlnet.ora.
>
> In the tnsnames.ora you can put
> (description =
> (sdu=NNNNN)
> ...
> )
>
> I can't remember off-hand where it goes in
> the listener entries - but I think it wasn't an
> exact match for the tnsnames entry.
>

Thanks, Jonathan. Our network requests are all fairly small (JDBC classes invoke PL/SQL programs with small arguments sent, less than 4k anyway), with the results sent back, so I couldn't figure out the issue, as small as it may be. It is odd, because there is no correlation between SQL*Net bytes from client, or any of the other SQL*Net type stats in v$sysstat. The only consistent correlation bewteen application response time and the Oracle wait interface is the "more data from client" event.

Also, another poster emailed me directly to clarify the numbers, and setting nodelay *decreased* response time from eighr centiseconds to four. My first post was incorrect (backwards), so my apologies.

Our next test will be swapping the JDBC jar files to see if that makes a difference, as it is still somewhat "jumpy".

Thanks again,

Steve Received on Sun Jan 28 2007 - 10:55:28 CST