Tomka,
Since this didn't post last time:
Another good test to do would be one that measures simultaenous steaming
read-write. For example:
create table_a, with at least 10,000 records, where field1 is the PK and
field2 is a text field.
create table_b, with 500,000 or more records, where field2 is an FK to
table_a, and field_3 is a text field.
do the following:
UPDATE table_b SET field3 = field3 || 'something'
WHERE EXISTS (select field1 FROM table_a
where table_a.field1 = table_b.field2
AND table_a.field_2 = 'x')
For the test to be most effective, the condition above should affect about
100,000 records of table_a, and the records should not be contiguos on disk.
Further, tablea.field2 should not be indexed in order to force a disk scan.
Also another test I'd really like to see is hardware raid (Adaptec, LSI)
against Linux software raid, and 5-disk RAID 5 against 4-disk RAID 1+0.
--
-Josh Berkus
Aglio Database Solutions
San Francisco
