Operating System - HP-UX
1832869 Members
5647 Online
110048 Solutions
New Discussion

I need of Best of you Read performance on the disk

 
SOLVED
Go to solution
gigiz
Valued Contributor

I need of Best of you Read performance on the disk

Hallo,
I have this scenario :
Cluster Siebel
2 rp7420 in cluster
Ram 19 Gb
8 CPU da 1000
Oracle 8.1.7 in archive mode
storage DMX 3000 in fibra in RAID 3+1
Power Path

The problem is the read perfomance is very slowly and this configuration when was on solaris was the much fastest one ....

14 REPLIES 14
Eric Antunes
Honored Contributor

Re: I need of Best of you Read performance on the disk

Hi Gigiz,

Please post the following outputs:

#sar -b 3 33

#sar -v 3 33

And since this is Oracle:

select (bbc.total_waits*100/(cg.value+dbg.value)) "Buff busy ratio"
from v$system_event bbc,
v$sysstat cg, v$sysstat dbg
where bbc.event='buffer busy waits'
and cg.name ='consistent gets'
and dbg.name='db block gets';

select round(100 * sum(getmisses) / sum(gets), 2) "Misses Pct (%)"
from v$rowcache
having sum(gets) > 1000;

select (1-(sum(getmisses)/sum(gets))) * 100 "Hit Ratio"
from v$rowcache;

select round(((1-(sum(decode(name, 'physical reads', value,0)) /
(sum(decode(name,'db block gets', value,0))+
(sum(decode(name, 'consistent gets', value, 0))))))*100),2)
|| '%' "Buffer Cache Hit Ratio (%)"
from v$sysstat;

select round(100 * sum(reloads) / sum(pins), 2) "Reloads Pct (%)"
from v$librarycache
having sum(pins) > 1000;

select pool, name, bytes/1024/1024 "Free Size in MB"
from v$sgastat
where name='free memory';

Best Regards,

Eric Antunes
Each and every day is a good day to learn.
Steven E. Protter
Exalted Contributor

Re: I need of Best of you Read performance on the disk

Shalom,

Raid 3+1 ??

Thats a new one.

Perhaps Raid 5 will be fasted for Raid because it has the most copies of all data around.

You might want to use a performance utility to collect some information and find the bottleneck.

http://www.hpux.ws/system.perf.sh

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
gigiz
Valued Contributor

Re: I need of Best of you Read performance on the disk

this is the output of command sar and the select of oracle
Zinky
Honored Contributor

Re: I need of Best of you Read performance on the disk

Peace and Clean Air, GigiZ,

Can you as well post: "sar -d 5 5" and vmstat 5 5 -- IF you are still having issues?

Also, how did you configure your Oracle storage? Filesystems or RAW? Also, did you stipe the DMX LUNs and what Volume Manager did you use?

Hakuna Matata

Favourite Toy:
AMD Athlon II X6 1090T 6-core, 16GB RAM, 12TB ZFS RAIDZ-2 Storage. Linux Centos 5.6 running KVM Hypervisor. Virtual Machines: Ubuntu, Mint, Solaris 10, Windows 7 Professional, Windows XP Pro, Windows Server 2008R2, DOS 6.22, OpenFiler
gigiz
Valued Contributor

Re: I need of Best of you Read performance on the disk

In attach another information ....
Zinky
Honored Contributor

Re: I need of Best of you Read performance on the disk

This early I can already tell you have an I/O problem. Your vmstat output shows you've many processes blocked on IO resources plus your sar -d output indeed shows disks that have queue's more than 0.

Can you also send "vgdisplay -v vgoradata" please to confirm if you've striping or not and if your LUN used for oraarch is on its own LUN? Also, can you check what your kernel max queue depth is and what is recommended for a DMX? I recall for EMCs -- it should be no less than 16.

kmtune|egrep "dbc|qdepth"

The above command will show caching and qdepth settings you have.

Solutions I can recommend this early:

1. MIgrate your oracle storage to striped for the Oracle datafiles.

2. Provide a non-striped/RAID10 LUN(s) for your oraarch storage so it is separate from your oracle datafile stripes.


Hakuna Matata

Favourite Toy:
AMD Athlon II X6 1090T 6-core, 16GB RAM, 12TB ZFS RAIDZ-2 Storage. Linux Centos 5.6 running KVM Hypervisor. Virtual Machines: Ubuntu, Mint, Solaris 10, Windows 7 Professional, Windows XP Pro, Windows Server 2008R2, DOS 6.22, OpenFiler
Eric Antunes
Honored Contributor

Re: I need of Best of you Read performance on the disk

Hi again,

From your stats, I recommend:

Increase dbc_max_pct to 10 because 62% is a very bad buffer cache ratio.

Decrease ninode to 2048 since you are not using those 392946.

Your Oracle stats seems ok...

Best Regards,

Eric Antunes
Each and every day is a good day to learn.
Eric Antunes
Honored Contributor

Re: I need of Best of you Read performance on the disk

You can do those kernel changes in SAM (you will need to reboot).

If you have any question about how to do this don't hesitate...

Best Regards,

Eric Antunes
Each and every day is a good day to learn.
gigiz
Valued Contributor

Re: I need of Best of you Read performance on the disk

in attach my vgdisplay, ktune, lvdisplay ....
gigiz
Valued Contributor

Re: I need of Best of you Read performance on the disk

in attach my vgdisplay, ktune, lvdisplay ....
Zinky
Honored Contributor

Re: I need of Best of you Read performance on the disk

One last info: see if "mount -p" shows your oradata mounted as directIO.

This early though, I can suggest a re-org of your storage:

1. stripe lvora 4 or 8 - way accross your DMX RAID3+1 LUNs.

2. For your oraarch volume.. leave it as a simple lvol (non-stripe) but better if you can get it to a RAID10 LUN instead of a RAID3+1 LUN.

3. If you do not have directIO enabled (,mincache=direct,convosync=direct) in your oradata mount .. please mount it using directio...

HTH.

Hakuna Matata

Favourite Toy:
AMD Athlon II X6 1090T 6-core, 16GB RAM, 12TB ZFS RAIDZ-2 Storage. Linux Centos 5.6 running KVM Hypervisor. Virtual Machines: Ubuntu, Mint, Solaris 10, Windows 7 Professional, Windows XP Pro, Windows Server 2008R2, DOS 6.22, OpenFiler
gigiz
Valued Contributor

Re: I need of Best of you Read performance on the disk

this my mount -p

mount -p
/dev/vg00/lvol3 / vxfs ioerror=nodisable,log,dev=40000003 0 1
/dev/vg00/lvol1 /stand hfs defaults,dev=40000001 0 0
/dev/vg00/lvol8 /var vxfs ioerror=mwdisable,delaylog,dev=40000008 0 0
/dev/vg00/lvol7 /usr vxfs ioerror=mwdisable,delaylog,dev=40000007 0 0
/dev/vg00/lvol6 /tmp vxfs ioerror=mwdisable,delaylog,dev=40000006 0 0
/dev/vg00/lvoraeng /software vxfs ioerror=mwdisable,delaylog,dev=40000009 0 0
/dev/vg00/lvol5 /opt vxfs ioerror=mwdisable,delaylog,dev=40000005 0 0
/dev/vg00/lvol4 /home vxfs ioerror=mwdisable,delaylog,dev=40000004 0 0
-hosts /net autofs ignore,indirect,nosuid,soft,nobrowse,dev=5f000000 0 0
/dev/vgoradata/lvoraarc /archive vxfs ioerror=mwdisable,delaylog,dev=40010001 0 0
/dev/vgoradata/lvorabck /backup vxfs ioerror=mwdisable,delaylog,dev=40010002 0 0
/dev/vgoradata/lvoradata /data vxfs ioerror=mwdisable,delaylog,dev=40010003 0 0
/dev/vgoradata/lvcdw /spare vxfs ioerror=mwdisable,delaylog,dev=40010004 0 0

what you want to say with stripe lvora 4 or 8 - way accross your DMX RAID3+1 LUNs.

And do you know kernel parameter for dmx 3000
Zinky
Honored Contributor
Solution

Re: I need of Best of you Read performance on the disk

GigiZ,

As I suspected.. you do not have the Oracle best opractive on Filesystems of having enabled DirectIO.

Please change your fstab entry for oradata so you have directIO:

/dev/vx/dsk/vgoradata/lvoradata /data vxfs rw,suid,largefiles,convosync=d
irect,mincache=direct,datainlog 0 2

Also, I would strongly urge you to recreate your lvoradata lvol so it is a stripe accross your DMX LUNs:

lvcreate -L 122880 -i 4 -I 64 -n lvoradata /dev/vgoradata /dev/dsk/c10t9d3... c10t9dd6


Also make oraarch on its own PV.

Hakuna Matata

Favourite Toy:
AMD Athlon II X6 1090T 6-core, 16GB RAM, 12TB ZFS RAIDZ-2 Storage. Linux Centos 5.6 running KVM Hypervisor. Virtual Machines: Ubuntu, Mint, Solaris 10, Windows 7 Professional, Windows XP Pro, Windows Server 2008R2, DOS 6.22, OpenFiler
gigiz
Valued Contributor

Re: I need of Best of you Read performance on the disk

Thanks Nelson ...
you aren't Caparroso ....
you are CAPA GROSSO that in Italian it wants to say you are the Best ....

:-)