Re: Weird Oracle 12.2 issue ..
Date: Fri, 15 Sep 2017 02:40:16 +0300
Message-ID: <CAOVevU6+0bvd0MMm77178msjgowzQG+qt0Re43A6dm5Db5Da3w_at_mail.gmail.com>
Hi Upendra,
First of all, that is not real execution plan, that's just the plan
produced by "autotrace". It uses explain plan really.
You may noticed that statistics showed 1168 SQL*Net roundtrips, though
you got just 583 rows. I suspect that your query contains LOB columns in
select list - it could cause extra roundtrips.
11 seconds/1168 roundtrips = 8.5ms per roundtrip - it doesn't seem like
something strange.
I think the easiest way to check it is to enable sql trace and give us raw
trace.
Also I'd set the following parameters:
set termout off arraysize 1000
and execute your query as script.
On Fri, Sep 15, 2017 at 2:17 AM, Upendra nerilla <nupendra_at_hotmail.com> wrote:
> All,
>
> We are running into a unique issue.. We recently migrated a database from
> 11.2 to 12.2 for testing.
>
> When we compared the performance, 12.2 performance was 3x-4x slow compared
> to 11g.
>
> When we were peeling the layers to troubleshoot, we discovered:
>
> - A query which runs in 3 seconds at the database directly (in 12.2)
> takes between 11-14 seconds when it is run through a client like pl/sql
> developer.
> - To eliminate the variables, we ran sqlplus from another host (which
> is in same network), and the execution time was still over 13 seconds.
> - sqlplus on the DB server (using tns/sqlnet) produces the result in
> about 3 seconds as well
> - Sysadmin says, Network throughput between the servers have no issue,
> he was able to scp a 500MB file in less than 2 seconds.
>
>
> Here is the output from SQL trace.. the output from running the query from
> the DB server and through the client produce identical output, except the
> elapsed time..
>
>
> <snip>
>
> 583 rows selected.
>
>
> *Elapsed: 00:00:02.51 vs Elapsed: 00:00:11.44 *
>
>
> Statistics
> ----------------------------------------------------------
> 66503 recursive calls
> 0 db block gets
> 71151 consistent gets
> 0 physical reads
> 0 redo size
> 382206 bytes sent via SQL*Net to client
> 248383 bytes received via SQL*Net from client
> 1168 SQL*Net roundtrips to/from client
> 0 sorts (memory)
> 0 sorts (disk)
> 583 rows processed
> <snip>
>
> Any suggestions on what I could check to pin-point the issue might be
> helpful..
>
> Thanks in advance
>
>
> -Upendra
>
-- Best regards, Sayan Malakshinov Oracle performance tuning engineer Oracle ACE Associate http://orasql.org -- http://www.freelists.org/webpage/oracle-lReceived on Fri Sep 15 2017 - 01:40:16 CEST