Re: Temp (sort) on Raid 5....

Not a lot of people know this, but many RAID-5 implementations these days can be tuned quite nicely for sort/temp.

The biggest complaint made about RAID-5 is the double write cost: you write the block and you read, recalculate, and rewrite the parity.

But if you do a full stripe write (i.e. the right N-1 discs), then the subsystem can calculate the totally new parity in memory and write the whole stripe with just a 1 in N overhead.

So with suitable configuration you can prove that RAID-5 is appalling, or RAID-5 if fine - what's your preference ?

To prove your point - write a piece of SQL that does a Cartesian sort-merge on about 10,000 rows, then groups the intermediate set to about 10,000 rows - this should be low on disc activity other than sorting. Set Sort_area_size to 512K, and watch v$session_event for 'direct write' waits.

--

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

Practical Oracle 8i:  Building Efficient Databases
Publishers:  Addison Wesley Longman
Book bound date now 1st Dec 2000

Doug wrote in message <1tgfvs4ohhigt9djuf0ueg0robs5asa2ca_at_4ax.com>...


>Souds like a horrible idea, doesn't it?   But, unfortunately I'm in

>the position of having to prove it.   Any ideas?  Any way to

>demonstrate with sysstat or system_event or something that sorting is

>running like glue?