Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Last week's drama

Last week's drama

From: Poorna Prakash <pooma.prakash_at_amrcorp.com>
Date: Sun, 05 Sep 1999 15:08:34 -0500
Message-ID: <37D2CDC2.FE7B4446@amrcorp.com>


Hi Bill,
In an effort not to miss, sharing with you, all of the interesting encounters
I had the past week, I had to record it in an email, so that we can address
them next week.

The good news....



Today, Sunday the 5th, I upgraded the SFA and Rockpile database to version 8.0.5. I also took a backup of the database .

Tuned the CMSDEV & CMSTEST database for performance.

The interesting news....



Last Tuesday, immediately after you left for vacation, CMSDEV database was rebooted for lack of performance. On wednesday I traced several user's sessions, and the database as a whole, and found interesting statistics. On Thursday the database performance slipped again and that's
when I decided to make changes to the database setup.

Summary of database performance statistics before enhancements:


  1. Excessive library cache misses was detected.
  2. Buffer cache hit ratio was only 70%.
  3. Processes were waiting for space in the redo log buffer
  4. The database was going through a lot ( I mean a lot) of full table scans. This was attributed to SAP Buffer table queries.
  5. The OS main memory was too low for PGA memory allocation, paging and swapping, as more than 60 users were logged on at one time.
  6. The total bytes of a single row in the SAP buffer table exceeds 2K, which eventually has caused all the rows to chain, and we all know what row chaining does to the database.

Here's what I did, to improve the performance and has worked :


  1. Decreased the CMSTEST instance's SGA size, in order to expand the size of CMSDEV's SGA.
  2. Increased the shared pool size
  3. Increased the database block buffer size
  4. Increased the log buffer size.
  5. Created indexes on the SAP buffer tables to avoid full table scans that were mostly occupied in the SGA at any one time.
  6. Gave more room to the OS for its swapping and paging.

Here's what I would like to suggest for future enhancements:


  1. We can increase the main memory from 256M to 512M, because CMSDEV is used by more number of users for 30 testing now, and we need a larger SGA.
  2. After the memory increase, we can expand the SGA size, to get all of the SGA hit ratios to 100%, in CMSDEV.
  3. We need to increase the default block size from 2K to 4K on all of the databases (not an easy task).

Sorry for this long email, but I wanted to communicate all the fun we had
last week.

-Poorna Prakash
 967-1043 Received on Sun Sep 05 1999 - 15:08:34 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US