Re: RAID-5 vs. Disk Mirroring

In <4ebtuu$9ln_at_maverick.tad.eds.com> bob_at_latcost1.alaao.ats.eds.com (Bob Stewart) writes:
>
>BBAP (bbap_at_aol.com) wrote:
>: Does anyone have experience to share on either RAID-5 or Disk Mirroring.
>: We are trying to make a decision on a direction. The concern is which
>: method will ensure that data will be there (no loss is permited), while
>: not comprimising performance. Cost is not a consideration.
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>

 You wrote:
>
>BBAP (bbap_at_aol.com) wrote:
>: Does anyone have experience to share on either RAID-5 or Disk Mirroring.
>: We are trying to make a decision on a direction. The concern is which
>: method will ensure that data will be there (no loss is permited), while
>: not comprimising performance. Cost is not a consideration.
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
>When do I start?
>
>Seriously. If cost is not really a consideration, then don't bother
>looking at RAID5. RAID5 solutions are considerably slower on average
>than mirroring solutions. Both are reliable.
>
>Later.
>--
>Bob Stewart ASE
>(310) 335-7152 Air Transportation Division
>bob_at_latcost1.alaao.ats.eds.com
>
>I am definitely NOT speaking for EDS.

In all of the discussions about RAID vs Disk Mirroring you need to understand that Oracle has 4 classes of files that have different I/O behavior if you match the RAID Level to the I/O behavior you will have an efficient and reliable system. Then again if price is no object then always use RAID 1/O as it gives the best throughput but at the highest cost 2x.

Also regarding the advise of SUN sysadmin on extending a RAID 1/0 system.

it should only be necessary to

1. shutdown the database.
2. copy the database files to tape
3. rebuild the file system to include the new drives
4. remount the file system to the same mount point
5. copy the database files from tape to disk
6. restart the database

remember that on some operating system the size of a file system is limited to 4gb and the size of a single file to 2gb. So you may need to create multiple file systems for a single Raid subsystem.

Kevin McHugh

File and Raid Comparison

1. Redo logs Synchronous Sequential writes of from 8 to 64kb
2. Archive logs Asynchronous Sequential writes of large (file system dependent) size
3. Tablespace data files Synchronous reads of by user processes of either 1 oracle block or multibyte read count blocks and Asynchronous writes by the database writer or writers of Oracle Blocks.
4. Tablespace for Rollback and Temporary segments. Synchronous reads of by user processes of either 1 oracle block or multibyte read count blocks and Synchronous writes by the database writer or writers of Oracle Blocks.

Match this with the Raid levels

RAID Level with N disks

0- Simple disk striping

    Seek 1x Transfer 1x Read and Write Throughput Nx i/o's per second     reliability (1/N) MTBF

1-Simple Mirror

    Seek x/N Transfer 1x Throughput Read Nx Write x/N i/o's per second     reliability MTBF**N
    assumes independent read access and pessimistic write (both I/O's     complete for write to return)

1/0 Stripped and Mirrored (S,N Disks)

    Seek x/N Transfer 1x Throughput Read (S*N)x Write x/N i/o's per second     reliability (MTBF**S)/N

3 Parallel Transfer Disks

    Seek 1x Transfer (N-1)x Throughput Read x Write x i/o's per second     reliability (MTBF**2)/((n**2-n)/2) Any 2 drives failing cause complete     failure.

5 Sector interleaved with crc sector.

    Seek 1x Transfer 1x Throughput Read Nx Write x/2 i/o's per second     reliability (MTBF**2)/((n**2-n)/2) Any 2 drives failing cause complete     failure.

1. Redo logs Simple mirror is best offers high reliabilty and good write performance.
2. Archive Log Simple mirror or 1/0 if volume of archives to buffer on disk is large.
3. Tablespace data files. For environments with high write volumes use 1/0 For predominantly read environments use 5
4. Tablespace data files RBS and Temporary Use 1/0 as both the reads and the writes are synchronous.