Performance - moving from oracle to postgresql - Mailing list pgsql-performance

From Greg Maples
Subject Performance - moving from oracle to postgresql
Date
Msg-id 2F9239F5E316C54688FDA9814FF045D6AB0024@EXCHANGE.sf.nimblefish.com
Whole thread Raw
Responses Re: Performance - moving from oracle to postgresql  (Rod Taylor <pg@rbt.ca>)
Re: Performance - moving from oracle to postgresql  ("Radu-Adrian Popescu" <radu.popescu@aldratech.com>)
List pgsql-performance
Hi:
    I'm beginning the push at our company to look at running
postgreSQL in production here.  We have a dual CPU 2.8 GHZ Xeon
Box running oracle.  Typical CPU load runs between 20% and 90%.
Raw DB size is about 200GB.  We hit the disk at roughly 15MB/s
read volume and 3MB/s write.
    At any given time we have from 2 to 70 sessions running
on the instance.  Sessions often persist for 24 hours or more.

                  Total      Free      Free
                  Mb           Mb          %

 IDXS_EXT10      2000        290      14.5
 DATA_EXT100      10000     3200      32
 SYSTEM            220           95.2      43.3
 IDXS_EXT100      20000     9600      48
 DATA_EXT10      6000           2990      49.8
 UNDOB            4000       2561.1 64
 TEMP           8000           5802.9 72.5
 DATA_LOB_EXT20      2000           1560      78
 IDXS_EXT1      500           401      80.2
 DATA_EXT1      4000           3758      94
Total Instance     56720     30258.2     53.3


There are some immediate questions from our engineers about performance

"- Oracle has one particular performance enhancement that Postgres is
missing.  If you do a select that returns 100,000 rows in a given order,
and all you want are rows 99101 to 99200, then Oracle can do that very
efficiently.  With Postgres, it has to read the first 99200 rows and
then discard the first 99100.  But...  If we really want to look at
performance, then we ought to put together a set of benchmarks of some
typical tasks."

Is this accurate:
accoring to
http://www.postgresql.org/docs/8.0/interactive/queries-limit.html
  -- " The rows skipped by an OFFSET clause still have to be computed
inside the server; therefore a large OFFSET can be inefficient."


What are the key performance areas I should be looking at?
Where is psql not appropriate to replace Oracle?

Thanks in advance, apologies if this occurs as spam, please send
Replies to me off-list.

pgsql-performance by date:

Previous
From: Rod Taylor
Date:
Subject: Re: max_connections / shared_buffers /
Next
From: Puddle
Date:
Subject: Re: max_connections / shared_buffers / effective_cache_size questions