Re: performance question
Date: 1996/06/24
Message-ID: <4qmim5$94t_at_nnrp1.news.primenet.com>#1/1
Willy Klotz <willyk_at_t-online.de> wrote:
: Terry Rose <terryr_at_inetnebr.com> wrote:
: >My company is currently evaluating Oracle for OLTP. On the server side, we have a
: >Tricord with 4 Pentium CPU's, 512MB RAM, running Solaris 2.5 and Oracle 7.3. On
: >the client side, we're currently using Windows NT 3.51.
: >I have an OCI application being used to benchmark Oracle performance. A table is
: >created with two columns: key CHAR(5), data CHAR(75). An index is created on the
: >key.
: >In the first phase, 20000 rows with unique key values 0-19999 are inserted in
: >random order. The data column is initially blank.
: >The second phase involves selecting 10000 rows by key, in random order.
: >My question is about the third phase which updates the data column of 10000 random
: >rows. This phase progresses in "bursts". Using a displayed counter, I can see
: >several hundred updates...then a pause...several hundred more...a pause...etc.
: >What is causing the pauses?
: >The final phase updates the key column of 10000 random row to 'ZZZZZ'. This phase
: >also exhibits the burst behavior.
: >Performance in all phases has been disappointing. We have tried adjusting the
: >server configuration from SMALL to LARGE. This resulted in better performance in
: >phase 2, worse performance in phases 3 and 4. Best times were recorded with
: >MEDIUM settings. Any other suggestions or tuning tips would be appreciated!
: I am very unsure that this type of "benchmark" will provide you with
: enough data to make a decision...
: The size of the tables you are playing around will easily fit into
: your RAM (if oracle is configured to let it in the SGA). If you want
: to do benchmarking, I would suggest that you create a set of more
: "live data" - I doubt that you will have this class of machine for a
: table with 20000 rows :-))
: Willy Klotz
: ======================================================================
: Willys Mail FidoNet 2:2474/117 2:2474/118
: Mailbox: analog 06297 910104
: ISDN 06297 910105
: Internet: 0629791010_at_t-online.de
: -> No Request from 06.00 to 08.00 <-
: ======================================================================
I agree with Willy's response. HOwever, if you want to use your original table for benchmarking, I would up the number of records to at least 500,000 rows and your other numbers should be increased proportionally.
As for the "Bursts" of activity, the pauses are due to writing of the redo logs the system pauses when it writes the logs to disk. I observed this same activity in a benchmark I did on HP equipment a couple of years ago.
If you have any other performance questions, please contact me directly. My company, Eagle's wing Software, specializes in Oracle performance.
-- ------------------------------------------------------------------------------- Donald Raikes CPIM Oracle Performance Specialist draikes_at_primenet.com /__)/__) / / / / /_ /\ / /_ / ------------------------------------------------------------------------------- / / \ / / / / /__ / \/ /___ /-------------------------------------------------------------------------------Received on Mon Jun 24 1996 - 00:00:00 CEST