Comments on: How does the sequence cache affect performance?

By: Asif Momen

Asif Momen — Sat, 01 Sep 2007 09:19:15 +0000

I have addressed a similar issue on my blog, have a look at it. .

By: Colin 't Hart

Colin 't Hart — Wed, 25 Oct 2006 09:25:13 +0000

Don’t forget that 1 REDO block will be created for each update of the sequence, which occurs every cachesize entries.

So if you are APPEND INSERTING 10s of millions of records and have a cachesize of 20 then you will still be generating quite a bit of REDO

Cheers,

Colin

By: Mathew Butler

Mathew Butler — Mon, 25 Sep 2006 13:37:56 +0000

Caching makes a significant different to batch data loads. Its a good tool to have when trying to tune a third party app without changing code – if the app is batch. I’ve found that a cache size of about 100 is optimal. Also coding the sequence directly into the DML cuts down the work to do. eg: insert into t ( col1 ) values ( my_seq.nextval ) returning col1 into l_my_col1; Of course, you can see greater benefits if the batch can be bulked up ie: using an insert into select …from…. Mathew Butler

By: Patrick Wolf

Patrick Wolf — Fri, 22 Sep 2006 16:10:19 +0000

Hi,

I think the cache can be very usefull if you have a RAC environment, because otherwise the different instances are playing ping-pong with the sequence dictionary tables all the time.

Patrick

By: Patrick Sinke

Patrick Sinke — Fri, 22 Sep 2006 13:28:53 +0000

I think there is an other optimal setting for each situations and each type of DML. My assumption was that cache size should be set just above the expected number of sequence calls within a large batch or procedure.
Testing some in-between values would yield a more reliable result of this test. Then we could also see if the relation is lineair or exponential, for instance.

By: Karl r.

Karl r. — Fri, 22 Sep 2006 12:54:24 +0000

hi,
interesting would be how good/bad the performance is if you cache 1000 Vals.
Sure the dictionary mut be updated a lot more but i am still interested in.

Karl