On 11 Sep 2013 07:24, Needles,Stephen J wrote:
<<SNIP>> We have a pretty good amount of memory and disk and are
not too concerned at this point about PAGESIZE, but will have to be
as we continue our trek. Is there a source to quantify the PAGESIZE
differences for planning purposes?
I am not aware of any specific docs, and my access to the InfoCenter
is problematic [so far] today. But basically, a simple arithmetic
representation can assist to help understand the effects. Given a key
size of 8-bytes, an 8K page size will have effectively 1K of key values
paged and accessible, but a 64K page size will have effectively 64K key
values paged. As /random access/ the key values being accessed from or
written to two different pages, results in 16K of paged memory in the
first example, but 128K of paged memory in the second. Of course 8X
memory requirements is maximal or worst-case with those two examples
[although I believe an 8-byte key is a 4K PagSiz], but given the nature
of keyed access and maintenance, the much larger page sizes as compared
to what would be typical\defaulted with DDS for RLA, could legitimately
significantly increase memory requirements.
"... IIRC a "short" key actually uses a 4K page size, and smaller page
sizes can be better for non-query [RLA] activity and effects on journal;
i.e. for effective memory footprint."