Re: [Info-Ingres] Time to say goodbye to the 2k page size?

At 8:05 AM -0700 10/24/07, Steve McElhinney wrote:
We will be upgrading several systems to Ingres 2006 R2 in the coming
months.

One proposal is that we start using 8k page size for *all* database
tables,
plus row-level locking on all tables.
i.e. this would be our default configuration for all Ing 2006 R2
systems
- unless for a particular application there is a good reason not to.

Any thoughts ... pitfalls? Has anyone gone down this route?

I'll chime in with another "yes, do it".


Datallegro appliances use a default page size of 32K, although
in a less specialized environment I think I'd go with a default
of 8K or just possibly 16K.

Watch out for tables with tiny rows, you're still subject to the
512 rows-per-page limit (or is it 511? I'm never sure...).
Small rows on a large page can waste a good bit of the page.
For the 8K page you're OK for pretty much any row width;
for (say) the 32K page, wastage starts at about 50+ bytes
per row.

If you go for the larger page sizes, 32k or 64k, watch out
for memory usage. DMF carries around lots of buffers that
you don't realize and it can add up at 32k or more.
(example: a table "load", which gets invoked by modify,
carries 16 one-page buffers, which at 32K is 512K of
load buffers. These are not in the DMF cache. Now,
modify a 5000-partition table ... that's 2.5 gig of memory.
The load buffer count is not configurable in standard Ingres.)

I'm not wildly enthused about row level locking as a default,
but that's more of an application specific thing. Bear in mind
that you won't get RLL in btree leaf pages unless you lower the
isolation level below serializable.

iidbdb (and I guess imadb) still gets the 2K page, and I've
never bothered to sysmod it to anything larger.

Karl
.