From: Gene Sais (Gsais@CO.PALM-BEACH.FL.US)
Date: Wed Nov 27 2002 - 09:55:25 EST
Oracle db_block_size can NOT be changed after the db is created prior to 9i. DB_BLOCK_SIZE is recommended to be at least a multiple of i/o block size.
hth,
Gene
>>> SGreen@KRAFTEUROPE.COM 11/25/02 07:19AM >>>
Your fr:sr ratio looks very high to me; I have only seen figures that high
in systems short of memory, although it's usually been accompanied by more
paging than you are seeing.
You have quite a high I/O Wait. How much of this is to your page spaces and
how much is normal I/O?
I am not aware of any document that describes how to set the Oracle
blocksize
for AIX. Generally, batch-type processing benefits more from high block
sizes
than on-line type, in my experience.
To get a better idea of memory utilisation, use svmon. Something like...
svmon -P -v -t 20 > /a/file
That'll show you the top 20 processes, sorted by the size of their virtual
storage, (i.e. paging space + real memory). You can try a larger number,
but
it can put quite a strain on the system if you get too high; not good on an
already busy system. Once you have this information you can try to get an
idea of how much space the biggest processes really need, and what
proportion
of that is typically paged out. From this, you can try to judge how much
more
memory would be required to keep more pages in memory. I've usually found
the
summary figures to be the most useful.
Since increasing minfree improved things, and you're not paging a great
deal,
I suspect that you don't need much more memory.
Simon Green
Philip Morris ITSC Europe
AIX-L Archive at http://marc.theaimsgroup.com/?l=aix-l&r=1&w=2
AIX FAQ at http://www.faqs.org/faqs/aix-faq/
N.B. Unsolicited email from vendors will seldom be appreciated.
-----Original Message-----
From: Holger.VanKoll@SWISSCOM.COM [mailto:Holger.VanKoll@SWISSCOM.COM]
Sent: 25 November 2002 11:14
To: aix-l@Princeton.EDU
Subject: performance problem - and 2 questions about oracle
Hello,
I got a s80 (4.3.3-10, 6cpu, 6gb ram) with 2 oracle-instances. One runs
batch-like (processing of large files), the other is interactively used by
users.
Those users complain about long response times.
Here is a vmstat 60:
2 6 1560288 706 0 1 0 4455 204014 0 2783 15595 7087 25 12 22 41
2 4 1560572 625 0 1 0 3228 152413 0 2084 13402 4684 29 9 33 28
<SNIP more vmstat output>
I dont have any system with such a high fr:sr ratio. Did anybody ever notice
such a high ratio?
I constantly see lrud in topas with 1-4% cpu; regarding this sr s I dont
expect anything else.
So I tried to remove some load from lrud and decreased lrubucket from 131072
to 65536. Users told me responsetime got a bit better.
Earlier I increased minfree from 120 to 600 (as I watched fre getting 0
sometimes), that also improved responsetime. But its still too bad.
As all memory-tuning improved responsetime, memory is (to me) the thing to
look at.
The big question: How can I determine how much memory would improve
responsetime?
2 things regarding oracle:
The db_blocksize is set to 8kb. Aix vmm writes pages, those are 4kb. I know
(and understand) the db blocksize should be equal to or multiples of the
"os-blocksize".
But is there any docu that tells when it should be equal and when multiple
times? You cant change this size after creating the db, so trial-and-error
is no way.
Does oracle support the aix-feature direct I/O?
http://publib.boulder.ibm.com/doc_link/en_US/a_doc_lib/aixbman/prftungd/2365
c813.htm
Here is vmtune -a output:
<SNIP>
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:16:23 EDT