Re: row level vs page level locking is it more than marketing hype? was Re: Informix vs. Sybase vs. Oracle vs. (gasp) MS SQL Server

Pablo Sanchez <pablo_at_sgi.com> wrote:

>>>>>> "Gary" == Gary Kuever <gkuever_at_ix.netcom_remove_this_.com> writes:
>Gary>
>Gary> 2 - On the minus side of row locking, it encourages developers to continue
>Gary> with a record oriented mentality instead of a set mentality. An example is
>Gary> Oracle's row_id. I've seen this on every Oracle project, i.e. the shortcut
>Gary> is taken instead of thinking the set operation through properly.
>Gary>
 

>... and of course you end up with crappier performance with
>the row at a time vs set based approach.
 

>I was doing a bench for a customer and they were doing some
>nightly processing... looking at their code, I saw that one
>section was taking 2.5 hours. I rewrote it and that same
>section took a couple of minutes.
 

>What was the magic? Converted their row at a time logic to
>set based logic.

Sure, that is the way to go. But in this set based logic it still would be nice if just the involved members of the set get locked when updating the set, and not all other members in the same table that just happen to be in the neighbourhood.
That is the point RLL fans try to make, not that they want to process 10,000 rows in a 1,000,000 row table one at a time. If I update all rows with an odd key, I do not want to block others that are only interested in even keys (for example).

I think we actually don't disagree, it's just that some people just want RLL because of their way of thinking (and it's that mindset that the PLL defenders are arguing against in this and the previous thread), but some others see valid arguments in favor of RLL in a well designed application (which IMO are not refuted by the PLL defenders in the current threads, better yet: which are agreed on).

In my opinion, to summarize the current status quo:

1. We all are against a row-at-a-time approach to application design
2. We all see PLL schemes perform better than RLL schemes (not going into the reasons why...)
3. We all see RLL advantages in that it gives a better relational/ setwise granularity with less unintended dependencies between rows in a table.

That will conclude my contribution to this thread (unless ... ;-)

Jean-Marc.

+------------------------------------------------------------+
|Jean-Marc.van.Leerdam_at_| All opinions expressed are just ... |
|ingbank.com           | opinions (and my personal ones!).   |