RE: Tuning in VXFS & Quick I/O

From: VIVEK_SHARMA <VIVEK_SHARMA_at_infy.com>
Date: Fri, 22 Jun 2001 08:26:03 -0700
Message-ID: <F001.00333C35.20010622083654@fatcity.com>

run iostat -xtc to monitor disks while benchmark is running. svc time will show any i/o bottle necks. Qs. - In iostat -xtc , Any VALUE of svc_t which if occurs can be Taken as Abnormally High ?
Try configuring multiple database writers. (Rule of thumb is 1 per cpu) Qs. - Since DISK_ASYNC_IO=TRUE , Configuring Multiple db_writers may NOT be required . The Best practices Doc. also Suggests Against Setting Multiple db_writers when DISK_ASYNC_IO=TRUE
Increase your SGA to decrease memory available to the OS. Qs. - SGA Can be Increased to what Percent of the Total RAM ? Of Now our SGA is 41% of Total RAM db_block_buffers=700M , shared_pool_size=90M , log_buffers=4M
Remeber that Oracle db_block_buffers acts as an I/O buffer. Qs. What Ratio Can this be increased by When Comparing with a NON-QIO Database ? We Have Increased to Double Now.

We have Reduced the db_file_direct_io_count to 8 from 64 (Default) as our Stripe Size is 64K
This is with the Belief that the I/O Buffer Size (=db_file_direct_io_count*db_block_size) should be LESS or EQUALto the Stripe Size.

Qs.

> -----Original Message-----
> From: Greg Connaughton [SMTP:greg.connaughton_at_corp.usa.net]
> Sent: Thursday, June 21, 2001 9:05 PM
> To: VIVEK_SHARMA
> Subject: Re: Tuning in VXFS & Quick I/O
>
> Vivek,
>
> Could be a lot of reasons Quick I/O isn't buying you much. You may
> have one
> device maxed out which is causing a bottle neck. Or you may not be
> doing
> enough I/O for quick I/O to buy you much. Where it really comes in
> handy is on
> a 8 - 10 cpu boxy with 8- 10 db writers. Quick I/O will allow multiple
> db
> writers can access the same file simultaneously where they can't as a
> regular
> FS. Also depending on how much memory you are using for your SGA,
> there may be
> plenty of memory for Solaris to cache disk I/O wich may be a short
> term boost.
> I don't know how long your benchmark is running. You might try the
> following.
>
> - run iostat -xtc to monitor disks while benchmark is running. svc
> time will
> show any i/o bottle necks.
> - Try configuring multiple database writers. (Rule of thumb is 1 per
> cpu)
> - Increase your SGA to decrease memory available to the OS.
> - Remeber that Oracle db_block_buffers acts as an I/O buffer.
>
> Good Luck!
>
> --
> Greg Connaughton Oracle DBA
> ***********
> Ph: 303.865.1243 USA.Net * *
> *
> Pager: 303.528.6170 7900 E.Union Ave *
> * *
> Fax: 303.865.1205 Suite 800 *
> * *
> email: Greg.Connaughton_at_corp.usa.net Denver, CO 80237-2735
> ***********
>
>
> > -----Original Message-----
> > From: VIVEK_SHARMA [SMTP:VIVEK_SHARMA_at_infy.com]
> > Sent: Wednesday, June 20, 2001 6:32 AM
> > To: LazyDBA.com Discussion
> > Subject: Tuning in VXFS & Quick I/O
> >
> > AIM - Performance Tuning on O.S. & Veritas
> >
> > SETUP -
> > DB Server :-
> > E3500 m/c = 2 CPUs , 2 GB RAM
> > Solaris 8 on ORA 8.1.7.1
> > Database on VXFS Filesystem with Quick I/O Installed .
> > RAID 0 with 64 K Stripe Size using VXVM & 8 K VXFS Block Size
> > Software RAID Configured as H/W RAID NOT Supported by the A5200
> Storage
> > Box.
> > db_block_size=8 K

Benchmarking Run Done on a Banking Application . 5000 Transactions Distributed Between 200 Concurrent User Processes are Executed
Transactions' Mix = OLTP & Batch Runs in the ratio 70:30 percent respectively

/etc/vfstab :-

/dev/vx/dsk/rootdg/vol01        /dev/vx/rdsk/rootdg/vol01       /in1/db1
vxfs    3       yes     mincache=direct,convosync=direct
/dev/vx/dsk/rootdg/vol02        /dev/vx/rdsk/rootdg/vol02       /in1/db2
vxfs    3       yes     mincache=direct,convosync=direct
/dev/vx/dsk/rootdg/vol03        /dev/vx/rdsk/rootdg/vol03       /in1/db3
vxfs    3       yes     mincache=direct,convosync=direct

/etc/vx/tunefs :-
/dev/vx/dsk/rootdg/vol01 qio_cache_enable=1

/dev/vx/dsk/rootdg/vol02 qio_cache_enable=1 /dev/vx/dsk/rootdg/vol03 qio_cache_enable=1

Qs. What Parameters Can be additionally Set on the O.S. & Veritas to Give a performance Benefit

Qs. Should the Tempfiles Also be Converted to Quick I/O (QIO) & Thereafter De-Sparsed ?

NOTE - After Converting the Existing Database's Files to Quick I/O (Running mkqio.sh) , NO performance benefit was Observed between the NON-QIO & QIO Converted Database in the Benchmark Runs .

Upon Setting the parameters mincache=direct,convosync=direct in /etc/vfstab & qio_cache_enable=1 in /etc/vx/tunefstab Files for the VXFS Filesyatems Containing the Database , Some Business Types of Transactions promarily Batch in Nature Improved Distinctly Though the OLTP Trans, Types Continued to perform at the Same Rate

--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: VIVEK_SHARMA
  INET: VIVEK_SHARMA_at_infy.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).

Received on Fri Jun 22 2001 - 10:26:03 CDT