Nurullah Akkaya wrote:
in my application i read data once compute and cache it myself it is
never read again so my initial thought was dropping pageCache to a
minimum ( 40 pages ) to save memory but since then i rewrite my query to
use only the index i am not hitting the tables at all. so my concern is
if i increase the pageCache to its original value or possible more does
it increase the performance of selects since i am not hitting the
table particularly one select i describe below. i do not want to mess
with the system before having an idea of its effects.
Anything anyone says is just going to be a guess for your application,
my suggestion is run a test representative of your application and try
it with 40 page cache, 400 page cache and 4000 page cache. It is
highly dependent on how likely each query is to access rows that were
previously accessed before they are flushed out of the cache. The
default cache might at least cache some of the high levels of the
index, but my guess is unless you cache the whole table the 2 million
row selects are mostly just going to be dominated by reading data
from disk - 1 million rows under 10 seconds seems pretty good to me.
table is around 30 gb 15 byte records one composite index includes all
columns that is the main index my application uses selects based on an
integer value. and it returns 100k to 2 million records. right now i can
read million rows under 10 seconds.
thx...
