We help IT Professionals succeed at work.

Dissapointing Performance after Migrating from dell R810 to R730 with Oracle Enterprise

We are upgrading an Oracle Enterprise server from a Dell R810 with 4 attached MD1200's (sas2 w/6g bus) to a Dell R730 with three attached MD1420's (sas3 w/12g bus).  R810 has four cores with 4 cpu threads and the R730 has two physical cores (2.6 GHz) with 8 way cpu's (3.2 GHz).  Drives on both systems are 15k.  Ram in each is 256gb (of course, the ram in the r730 is faster).

Problem is that we're only seeing 10-15% performance lift in the new server.  We expected much more. Lastly, the drives in both machines are SED.

Any ideas on getting better Oracle performance?

Thanks--
Comment
Watch Question

CERTIFIED EXPERT

Commented:
Nothing specific about Oracle, but when there are performance issues I look at Resource Monitor to see what is being maxed out.  This assumes you are running Windows.  Other OSs should have something similar.

If you do simple file transfers how does that speed compare to what you had in the old server?

Author

Commented:
It's a Solaris 11 box.  The cpu shows low contention.  Normaly, disk i/o has been the biggest performance variable in our systems, so I thought the faster sas 3 / 12g speed would be really noticeable. Can't move to ssd's due to encryption requirements and super high costs.
CERTIFIED EXPERT
Distinguished Expert 2019

Commented:
You have fewer physical disks before but it actually runs a bit faster? I'd be very happy with that.

Author

Commented:
Andy, the old MD1200's have 12 x 3.5" drives and the new MD1420 have 24 x 2.5" drives.  So, I've attached a toal of 48 drives to the old system and 72 drives to the new system.
CERTIFIED EXPERT
Distinguished Expert 2019

Commented:
did you follow the tunning of the system kernel, as well as

How big are the DBs?

Are the MDs on the new chained over the same BUS or you have multi-path/<ulti-channel?

how many external Perc H8xx do you have in the R730 versus.

two processor dual core with hyperthreading enabled.

your CPU breakdown is unclear " R810 has four cores with 4 cpu threads and the R730 has two physical cores (2.6 GHz) with 8 way cpu's (3.2 GHz)"
in whcih memory configuration did you setup the new server, optimal,. mirrored? multi-channel?


look at iostat, vmstat, to see where the bottleneck is.

consider the demand on data is going through a perc8xx which handles both RAID of the

are you using both controlling modules to connect/chain the external enclosures.

does the new setup matches replicates the old one? topology./arrangement?
CERTIFIED EXPERT
Top Expert 2008
Commented:
The performance issues begins with announcing the Oracle version.
E.g. 12,2,0.1 has a generic defect that can be fixed.


You have also to check if Solaris uses Transparent Huge Pages. If yes - immediately switch them off.

Next point is to check the SGA disposition.

Some queries for this are below.

Pay attention to the whole SGA size and the sizes of the following components:

- db_buffer.... size
-shared_pool ... size



############################ Current values ################################

   SELECT pool, round(sum(bytes)/1024/1024) Mbytes FROM v$sgastat WHERE pool is not null group by pool  
   UNION
   SELECT 'db_cache', current_size Mbytes FROM v$buffer_pool;



http://www.orafaq.com/wiki/SGA

################# current values, resizable, free #####################################

   set pagesize 1000

   SELECT name, round(bytes/1024/1024) MB, resizeable FROM v$sgainfo;


  SELECT name, value
  FROM   v$parameter
  WHERE  name in ('__shared_pool_size', '__java_pool_size',
 'streams_pool_size', 'log_buffer', '__db_cache_size', 'db_2k_cache_size',
 'db_4k_cache_size', 'db_8k_cache_size', 'db_16k_cache_size',
 'db_32k_cache_size', 'db_keep_cache_size', 'db_recycle_cache_size',
 '__large_pool_size');


################################# Detail statistics #########################
When automatic shared memory management is enabled, Oracle will adjust the memory parameters on the fly.
To see currently allocated sizes:
 
 set linesize 1000
 set pagesize 1000
 COLUMN component FORMAT A30

 SELECT * FROM v$sga_dynamic_components;

 set linesize 1000
 set pagesize 1000
 COLUMN component FORMAT A30

 SELECT component, current_size/1024/1024, min_size/1024/1024, user_specified_size/1024/1024 FROM v$sga_dynamic_components;
 

#######################################  SGA   #############################################
CERTIFIED EXPERT
Distinguished Expert 2019

Commented:
Sorry I was wrong about the number of disks so it must be some tuning issue, however if you are using Dell PERCs you can still use a few SED SSDs as read cache using CacheCade.