Re: buffer advisor (has become: should there be a private sql area?)

The design choice is not limited to converting versus flooding; you leave out the alternative: “versus not searching the pool, not using latches, and just parsing it privately.”

When parsing is a significant part of the total query cost and there is a high fan in, sharing is essential to scalability and mapping tends to work.

When parsing is an insignificant part of the total query cost sharing is not relevant to performance.

When no similar query is likely to be used, sharing is probably just overhead. Likewise, you don’t define outlines for things that have not be run before.

When sharing drives an avoidable error, that is the ultimate performance penalty.

Since private parsing has not been available since 6 (when it was the only choice), it is correct to ignore it as a trouble-shooting mechanism. It is not irrelevant in the context of an enhancement request to avoid shared pool contention and space errors or reduce the time to parse of simple queries containing literals. MANY literal queries retrieve one row by an obvious one row choice index. IF plan creation started with the heuristic: Is there an obvious plan that delivers cost < epsilon or cost < psi where epsilon is some multiple of the cost to look for a better plan and psi is the cost to search the shared pool, then avoiding sharing could be a gated scaling benefit.

Flooding the shared pool with un-sharable SQL is clearly a bad idea.

And sometimes you do not.   The question becomes would automatically converting constants to bind variables in all cases be more beneficial to the overall system verse flooding the shared pool with un-sharable SQL?  In the case where the code runs better with a constant then there are Outlines and SQL Profiles to assist with tuning.  Then again if Oracle works this way bind variable peeking might just peek on every execution for statements where it might make a difference.

Because sometimes you get a superior plan with a literal (by a lot),  and then we’d have to hint one way or the other and we cannot do that because sometimes (often) we do not control the source code.

Why not automatically convert all constants to bind variables as part of the parse?  This would make nearly identical SQL statements where only the value of the constant changes into identical statements supporting sharing.

There is no question that reusing reusable sql parses was a huge advance for scalability. But tossing non-reusable sql in the shared pool just puts extra pressure on the latches.

mwf

The shared pool came at the same time as database server  pl/sql  --- if I remember correctly.   [there was forms pl/sql available earlier]

Was there co-development or dependency ?

In fact, why not parse all sqls with literals privately? It worked just fine in V6.

mwf

This email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please delete all copies and notify the sender immediately. You may wish to refer to the incorporation details of Standard Chartered PLC, Standard Chartered Bank and their subsidiaries at https://www.sc.com/en/incorporation-details.html.