Sergei,
Thank you for the nice help.
> Try to experiment with bulk_insert_buffer variable. You need both big
> keycache and big bulk_insert_buffer.
I lowered key_buffer_size from 512M to 256M and increased bulk_insert_buffer_size
from 8M to 128M.
> When MERGE tables will support fulltext searches you can use them to
> keep small table where you insert data, and from time to time move the
> data from it to the big one. Actually you can do it with UNION even now,
> but it's not nice solution at all.
Oh, I will do anything to make it work .. I will do unions between 50 selects
if I must. I access MySQL via an abstracted perl api which I dont change,
so I can live with awful things underneath if there is just a shiny surface. If
unionizing doesnt slow fulltext search much, then I try that first.
Will a future MySQL version combine the current nice and fast word-search
with ability to add data in finite time? I saw "Make all operations with FULLTEXT
index faster" on the TODO list .. are you able to say what is in the pipe?
Greetings, Niels L
Bioinformatics Research Center (BIRC)
Aarhus University
Hoegh Guldbergsgade 10
DK 8000 Aarhus C
Denmark
