Re: parallel query

In article <3h88m3$7g8_at_jac.zko.dec.com>

           roderick_at_eps.enet.dec.com "Lisa Roderick" writes:

: > How can I use the paralle query option for this query.
:
: Do you have an SMP or clustered system? If not, parallel query won't help
: and can hurt performance.

I'm not entirely convinced about this, but I'm not an O/S specialist, so would someone comment on the sense or stupidity of the following argument:

	Scenario:		Single CPU
	Table:			Spread across several discs
	Degree of parallelism	Matches number of discs

Assume that there are 10 processes competing for CPU, one of which is your dedicated tablescan process: with a time-slicing scheduler, your process gets about 1/10 of a second per second.

Assume now that Oracle despatches 3 parallel scanners for you, ignoring your base process and query co-ordinator there are now 12 processes competing for CPU, but 3 of them are yours, so your tablescan gets 3/12 = 1/4 second of CPU per elapsed second.

Net result: everyone else slows down by 20%, you go 2.5 times as fast.

Where's the flaw ? Is the model _too_ simplistic to have any meaning at all ? I await your comments.

-- 
Jonathan Lewis