Oracle VM 2.2 :: Oracle Clusterware/Oracle 11gR2 {Single RAC node} Question

Question

1. Is it recommended to use physical disks everywher for ORAHOME, base operating system root, OCR / Vote disks and ASM disks? At this point of time in my vm.cfg only the ASM disks are physical while the OCR / Vote disks and ORAHOME and operating system is on an image file using tap:aio (blktap) driver.
Current vm.cfg disk directive
disk = ['tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/running_pool/16_oeldb1/System.img,xvda,w',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/running_pool/16_oeldb1/oracle.img,xvdb,w',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/oracle_oeldb2.img,xvdc,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/asm_vote01.img,xvdd,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/asm_vote02.img,xvde,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/asm_vote03.img,xvdf,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/asm_vote04.img,xvdg,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/asm_vote05.img,xvdh,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/db_disk_01.img,xvdi,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/db_disk_02.img,xvdj,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/db_disk_03.img,xvdk,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/db_disk_04.img,xvdl,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/oeldb_ocfs2.img,xvdm,w!',
'tap:aio:/var/ovs/mount/1FB2E6B36D21418792A1DD30B62C689F/sharedDisk/temp.img,xvdn,w!',
'phy:/dev/sdf,xvdo,w!',
'phy:/dev/sdg,xvdp,w!',
'phy:/dev/sdh,xvdq,w!',
'phy:/dev/sdi,xvdr,w!',
2. is it recommended to use OCFS2 base partitioned disks be used to create ASM disks. I want to use these disks for ASM. (OR should i destroy the filesystem at the hypervisor and just present raw disk)
[root@oeldb1 ~]# mounted.ocfs2 -d
Device FS UUID Label
/dev/xvdo ocfs2 67728d4c-f26d-4ec7-9ca1-4b25be453ec8 phydb1
/dev/xvdp ocfs2 6901ffd5-33cb-46f7-99bb-74fc712ea4db phydb2
/dev/xvdq ocfs2 95ed1075-f772-4c69-9ae0-9078b54bc8b5 phydb3
/dev/xvdr ocfs2 6b3c313e-cec0-4967-a670-8856d38b0cd1 phydb4
[root@oeldb1 ~]#

Answer 1

Sebastian : Thanks. I did come across this white paper & now I have converted they ASM disks to physical. When you say just direcly pass LUNs from storage, its still going to be a virtual block device off the RAW lun to OVS, is that what you meant? .. ie RAW disk to OVS --> use phy: directive to present and xvd"n" to vm (PV block driver)
ssolbach wrote:
Hi,
1.) It is recommended to use physical disks for all ASM disks (also for OCR and Vote). If you use 11.2. then OCR/Vote are on ASM anyway.
See also: http://www.oracle.com/technetwork/database/clustering/oracle-rac-in-oracle-vm-environment-131948.pdf
2.) Just present RAW disks. Best is to directly pass the LUNs provided from the storage.
Regards
SebastianAvi : hdparam and dd tests do not say so. I have seen better performances with blktap driver. Even iostat run within ASM showed better response times. I serched on white paper and documents within metalink on blktap or file driver, but couldnt find any. (I did find some ex Oracle SE's journals recommending blktap driver for better performance). Are you referring to significant improvements in 3.0.2 ? because, in 2.2 the file: driver is poor performing.
Avi Miller wrote:
Also note that you should be using the file: driver with Oracle VM, not tap:aio -- all of the Oracle performance improvements went into the loopback driver that uses the file method.Edited by: Abhijit on Dec 8, 2011 7:14 AM
Edited by: Abhijit on Dec 8, 2011 7:15 AM
Edited by: Abhijit on Dec 8, 2011 7:18 AM
Edited by: Abhijit on Dec 8, 2011 7:19 AM