There are two really big problems about the b-tree scanner:

1. It's mechanisms are not documented (Neil, thank You for your efforts
to compensate it)
2. The default configuration parameters are not suitable for any
real-world database: they cause infinite index page laundering and
replace any useful information in the Informix buffer cache with pages
of a single index currently being cleaned.

The fact is that, unlike old-style b-tree cleaner, b-tree scanner
MUST be properly configured - the effect of using the default
configuration parameters is comparable with leaving the 'BUFFERS' to
it's value from 'onconfig.std'

------------------------------------------
Alexey Sonkin


> From: owner-informix-list@iiug.org
[mailtowner-informix-list@iiug.org]
> On Behalf Of Michael Mueller
>
> Alexey Sonkin wrote:
> > Just put
> >
> > BTSCANNER num=1,priority=low,threshold=50000,rangesize=10000
> >
> > in Your $ONCONFIG - and Your desperation will go away
>
> My desperation might turn into sadness, no more that that. In my
opinion
> the new btree scanner has at least two major design flaws (I'm not
> saying that the old concept had none):
>
> 1) It has no idea which pages to clean and wastes a lot of i/o
bandwidth
> to find dirty pages by scanning whole indexes or at least whole
physical
> ranges of pages.
>
> 2) In some cases instant btree cleaning is a precondition for good
> insert performance. The new concept cannot deliver that.
>
> I would appreciate a discussion about this on this list.
>
> Michael


sending to informix-list