ASM Basics

Question

I have Oracle 10gR2 software installed in one machine and all datafiles kept in ASM disk group. Can anyone explain the physical side(like Hub, SCSI controller,..) of how the machine which has the Oracle software installed is connected to the ASM Disk group?

Answer 1

Hi,
Your question is quick to ask, but has a complex answer. You need to have a really good "hardware" and computer design knowledge..
I'll try to explain it shortly and simply.
There are different components in your setup.
1- physical disks: There you can have local storage (disks, raid array, ...) or shared storage (SAN, NAS,...). The computer on which you're going to run ASM must have physical access to the disk devices. He must see the LUNs if you're using a SAN, or he must see the raid array, or ...
2- the host(s): It must be configured to use ASM. On Unix platforms, the raw devices must be owned by the software owner (oracle), belong to the dba group (dba) and be in chmod 660. The SAN/RAID/... drivers must be loaded into kernel.
3- ASM: ASM is a filesystem. The host O.S. can't see what's in this kind of filesystem (except with Linux/asmlib and it's not so straightforward). Only Oracle binaries can use ASM filesystems. When ASM sees a raw disk which it can access (owned by oracle:dba mode 660 on unix, remember?) it reads the disk header to see what'sin the disk.
4- the diskgroups: they are made of different "disks" (1). When you tell ASM to create a diskgroup, it writes some information at the beginning of the "disk" in order to mark it as an ASM "disk". Then ASM manages redundancy, ..., as you told it it sould be done.
A little sample:
Disk1\                         / PARTITION 1 \
Disk2\\                        / PARTITION 2 - ASM DISKGROUP 1 \
Disk3--- RAID 5 ARRAY <-> HOST - PARTITION 3 \                 - ASM Instance <-> DB Instance
Disk4//                        \ PARTITION 4 - ASM DISKGROUP 2 /
Disk5/In the sample I quickly graphed, you can see there are 5 physical disks in a RAID 5 array. A Raid controller would certainly manage the array, and the controller would show one volume to the OS. It's possible to have a software raid manager (in linux for example) and in that case, the host O.S. is the raid controller.
At O.S. level, the raid volume is split into partitions, or can be used as-is, and the ASM instance is configures to use a set of partitions and organize them into disk groups. the ASM instance is the filesystem manager, the DB processes access this filesystem in order to read the files as an other software would read an NTFS volume via O.S. interfaces.
To give a more logical sample, here's the setup I have for my main standby database:
Disk1 \
Disk2 \        - RawDisk1a
Disk3  - SAN 1 - RawDisk2a
Diskn /
                             HOST <-> ASM DISKGROUP (RawDisk1a,RawDisk2a mirror on RawDisk1b,RawDisk2b) <-> ASM Instance <-> DB Instance
Disk1 \
Disk2 \        - RawDisk1b
Disk3  - SAN 2 - RawDisk2b
Diskn /Here the DB Instance uses the ASM filesystem to access a data diskgroup. The diskgroup is mirored on 2 SANs via ASM FAILGROUPs usage. ASM sees the RawDisk(n) via kernel API, but writes directly to it. The ASM software if in charge of the mirroring at ASM level (cf DISKGROUP REDUNDANCY LEVEL in the doc). Each RawDisk is set in a different failgroup, so if one SAN fails, ASM can still work using the other SAN.. If a disk in either SAN fails, ASM isn't even informed.. The SAN is in charge of the RAID redundancy.
Well, it's a complex matter.
I hope I cleared some of your doubts, and I'm sure I raised some more concerns.
Ask away if you want some more explanations.
Regards,
Yoann.