Comments on: Sequences in Oracle 10g RAC

By: Christo Kutrovsky

Christo Kutrovsky — Sun, 24 Jul 2011 13:45:57 +0000

Subrata, what does the official Oracle Documentation say about this?

By: subrata saha

subrata saha — Thu, 21 Jul 2011 23:11:28 +0000

cache+noredered option in RAC – does it gurantee order within an instance of RAC according to call to this Sequence

regards
subrata saha

By: Christo Kutrovsky

Christo Kutrovsky — Mon, 04 Apr 2011 19:35:26 +0000

Hi TalkingHead,

I think it’s correct as written. What I mean is that even “noordered” sequence can have a significant overhead if they use the default cache size.

By: TalkingHead

TalkingHead — Thu, 31 Mar 2011 20:42:32 +0000

I think there is a typo in the sentence below
“With a smaller cache, the “noordered” still has as significant impact as every 10 fetches (cache 20 divided by 2 nodes fetching) it has to synchronize between the 2 nodes”

Did you mean to say “ordered” instead of “noordered” ?

By: Oracle Performance: DFS Lock Handle and Oracle Sequence | twit88.com

Oracle Performance: DFS Lock Handle and Oracle Sequence | twit88.com — Sat, 29 Aug 2009 05:42:36 +0000

[...] http://www.pythian.com/news/383/sequences-in-oracle-10g-rac [...]

By: Rakesh

Rakesh — Mon, 27 Jul 2009 02:04:24 +0000

Hi,
Can any one tell me the procedure to change the cache size and sequence type? Is there any way to determine the cache size for a particular database? I know I am asking silly questions but I didn’t find appropriate material? A link to appropriate material will also be helpful?

Thanking you

By: Larry T

Larry T — Wed, 09 Apr 2008 13:38:11 +0000

I have always approached a SEQUENCE OBJECT as simply a way to guarantee uniqueness for a surrogate key – nothing else. Once you start adding expectations on SEQUENCEs that go beyond that, it will add overhead, complexity and risk.

By: Christo Kutrovsky

Christo Kutrovsky — Thu, 13 Sep 2007 14:35:27 +0000

average925joe:

Cached sequence values are lost in an instance crash. They can also be lost during pressure for memory in the shared pool. You should not rely on not loosing numbers.

IF you need to have a no-gap type sequence, you will need to pre-populate a table and manage it there.
With a sequence, you could loose numbers simply by requesting the next number, and rolling back after that. So even a nocache can “loose” numbers in that sense.

By: Francisco H.

Francisco H. — Tue, 14 Aug 2007 17:00:31 +0000

Great finding. However, our results draw a different conclusion. If you increase the loop size to 500,000 the size of the cache has no effect from 2000 to 20000. Performance is aweful when ordered.

Type Cache Order Loop size Time in Sec
1N 20 N 50,000 2.6
2N 20 N 50,000 11
1N 20 Y 50,000 3.3
2N 20 Y 50,000 28
2N 2000 Y 50,000 13
2N 20000 Y 50,000 11
2N 20000 Y 500,000 180
2N 2000 Y 500,000 160
1N 2000 Y 500,000 21
2N 2000 N 500,000 13
2N 20000 N 500,000 13
1N 20000 N 500,000 14

1N One Node
2N Two Node Simultaneous runs

By: average925joe

average925joe — Tue, 07 Aug 2007 07:35:25 +0000

Hi Chris, what happens to sequence values cached in SGA at the time of an instance crash (with cache option)? I read somewhere that you would loose values (sequence gap) but does that apply to a database restart or an instance crash? how can sequences be catered through the application (pre-generated) or using dbms_random or creating a procedure ? would appreciate your reply. Thanks

Cheap Kamagra