Re: avoiding corrupted block (ora-01578)

Coments interleaved:

--

Jonathan Lewis
Yet another Oracle-related web site: www.jlcomp.demon.co.uk

Oracle DBA (de-munge for real address) wrote in message <376a61c7.2067557210_at_news.connectnet.com>...

>The specific performance problem with RAID5 is that it incurrs
>a one for one overhead write operation for ever data block
>being written. When a datablock gets written to a RAID5
>volume the corresponding XOR block also needs to be updated
>or re-written.

    It can be worse than that - it may be necessary to read     the XOR block before it can be updated

>For example it would be a VERY POOR choice to place the
>archive log files on a RAID5 volume; because these files
>only get written to. Effectively the throughput would be
>reduced by about 50% to archive logs files on RAID5.

    Good example of where 'everyone knows it' but its not     necessarily true. Archived redo can be configured to     write in large stripes (typical is 64K, some do more).

    Some RAID5 drivers can recognise when the device     is being asked to write a full stripe, and calculate the     XOR block once from the incoming data without a     read. The XOR block is then just the 'N+1' block     write, not a block per block 'read and write'. Received on Fri Jun 18 1999 - 11:02:08 CDT