Comments on: What happens, when you did not do a load test before release upgrades ?

By: Lei Zeng

Lei Zeng — Wed, 16 May 2012 22:12:50 +0000

Thank you for your quick reply. I noticed that there are two “SQL Stats -Pre-11G”, is that a typo? Do you really mean one is for “11g” and one is “pre 11g”?
Thanks again

“

By: coskan

coskan — Wed, 16 May 2012 19:02:04 +0000

it is there in v$sql table

select
inst_id ins,
child_number sql_child_number,
IO_CELL_OFFLOAD_ELIGIBLE_BYTES/1024/1024/1024 io_elig,
IO_INTERCONNECT_BYTES/1024/1024/1024 io_inter,
child_address object_handle,
users_executing uex,
plan_hash_value plan_hash,
OPTIMIZER_COST opt_cost,
parse_calls parses,
loads h_parses,
executions,
fetches,
rows_processed,
cpu_time/1000 cpu_ms,
elapsed_time/1000 ela_ms,
(elapsed_time/1000000)/decode(nvl(executions,0),0,1,executions) avg_ela,
buffer_gets LIOS,
disk_reads PIOS,
sorts,
locked_total,
pinned_total,
is_obsolete is_o,
is_bind_sensitive is_bs,
is_bind_aware is_ba,
is_shareable is_sh,
– address,
– sharable_mem,
– persistent_mem,
– runtime_mem,
optimizer_env_hash_value opt_env_hash
from
gv$sql
where
sql_id = (‘&1′)
order BY
inst_id,
sql_id,
hash_value,
child_number
/

By: Lei Zeng

Lei Zeng — Wed, 16 May 2012 18:56:54 +0000

Would you please share you script ‘sqlid’.

@sqlid

I am wondering how it pull out PARSES and H_PARSES information.

Thank you very much.

By: LisaG

LisaG — Wed, 17 Aug 2011 10:23:24 +0000

We have seen similar issues in Oim -Oracle identity manager. This one hammered the san over 30 threads. Oracle answer was to purge tables.
OEM was the second application and the answer was a hint in the code.
I can see how this is an awesome feature on exadata , but it hammers a shared san infrastructure.

By: Dom Brooks

Dom Brooks — Tue, 16 Aug 2011 09:32:17 +0000

Another great article, Coskan.
A third solution – changing _small_table_threshold,as mentioned in some of the other articles, either directly or indirectly (increasing sga),

We’re currently working through an upgrade which includes running batch workloads multiple times, changing config, flashing back, rerunning, etc.

New oracle version means new box means bigger memory.

As we’ve increased the memory allocation from our initial tests on current settings, we’ve seen a two-pronged benefit from a) obviously a bigger buffer cache and less physical IO but also b) less serial direct path reads as the bigger allocation means a higher small_table_threshold.

It’s not a feature that I’m convinced about. I’ve only noticed performance problems associated with it in this upgrade so far – but then again maybe the benefits have silently passed me by?